Loading...
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
16static int __remove_xattr(struct ceph_inode_info *ci,
17 struct ceph_inode_xattr *xattr);
18
19static const struct xattr_handler ceph_other_xattr_handler;
20
21/*
22 * List of handlers for synthetic system.* attributes. Other
23 * attributes are handled directly.
24 */
25const struct xattr_handler *ceph_xattr_handlers[] = {
26#ifdef CONFIG_CEPH_FS_POSIX_ACL
27 &posix_acl_access_xattr_handler,
28 &posix_acl_default_xattr_handler,
29#endif
30 &ceph_other_xattr_handler,
31 NULL,
32};
33
34static bool ceph_is_valid_xattr(const char *name)
35{
36 return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
37 !strncmp(name, XATTR_SECURITY_PREFIX,
38 XATTR_SECURITY_PREFIX_LEN) ||
39 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
40 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
41}
42
43/*
44 * These define virtual xattrs exposing the recursive directory
45 * statistics and layout metadata.
46 */
47struct ceph_vxattr {
48 char *name;
49 size_t name_size; /* strlen(name) + 1 (for '\0') */
50 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
51 size_t size);
52 bool readonly, hidden;
53 bool (*exists_cb)(struct ceph_inode_info *ci);
54};
55
56/* layouts */
57
58static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
59{
60 struct ceph_file_layout *fl = &ci->i_layout;
61 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
62 fl->object_size > 0 || fl->pool_id >= 0 ||
63 rcu_dereference_raw(fl->pool_ns) != NULL);
64}
65
66static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
67 size_t size)
68{
69 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
70 struct ceph_osd_client *osdc = &fsc->client->osdc;
71 struct ceph_string *pool_ns;
72 s64 pool = ci->i_layout.pool_id;
73 const char *pool_name;
74 const char *ns_field = " pool_namespace=";
75 char buf[128];
76 size_t len, total_len = 0;
77 int ret;
78
79 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
80
81 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
82 down_read(&osdc->lock);
83 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
84 if (pool_name) {
85 len = snprintf(buf, sizeof(buf),
86 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
87 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
88 ci->i_layout.object_size);
89 total_len = len + strlen(pool_name);
90 } else {
91 len = snprintf(buf, sizeof(buf),
92 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
93 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
94 ci->i_layout.object_size, (unsigned long long)pool);
95 total_len = len;
96 }
97
98 if (pool_ns)
99 total_len += strlen(ns_field) + pool_ns->len;
100
101 if (!size) {
102 ret = total_len;
103 } else if (total_len > size) {
104 ret = -ERANGE;
105 } else {
106 memcpy(val, buf, len);
107 ret = len;
108 if (pool_name) {
109 len = strlen(pool_name);
110 memcpy(val + ret, pool_name, len);
111 ret += len;
112 }
113 if (pool_ns) {
114 len = strlen(ns_field);
115 memcpy(val + ret, ns_field, len);
116 ret += len;
117 memcpy(val + ret, pool_ns->str, pool_ns->len);
118 ret += pool_ns->len;
119 }
120 }
121 up_read(&osdc->lock);
122 ceph_put_string(pool_ns);
123 return ret;
124}
125
126static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
127 char *val, size_t size)
128{
129 return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
130}
131
132static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
133 char *val, size_t size)
134{
135 return snprintf(val, size, "%u", ci->i_layout.stripe_count);
136}
137
138static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
139 char *val, size_t size)
140{
141 return snprintf(val, size, "%u", ci->i_layout.object_size);
142}
143
144static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
145 char *val, size_t size)
146{
147 int ret;
148 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
149 struct ceph_osd_client *osdc = &fsc->client->osdc;
150 s64 pool = ci->i_layout.pool_id;
151 const char *pool_name;
152
153 down_read(&osdc->lock);
154 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
155 if (pool_name)
156 ret = snprintf(val, size, "%s", pool_name);
157 else
158 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
159 up_read(&osdc->lock);
160 return ret;
161}
162
163static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
164 char *val, size_t size)
165{
166 int ret = 0;
167 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
168 if (ns) {
169 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
170 ceph_put_string(ns);
171 }
172 return ret;
173}
174
175/* directories */
176
177static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
178 size_t size)
179{
180 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
181}
182
183static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
184 size_t size)
185{
186 return snprintf(val, size, "%lld", ci->i_files);
187}
188
189static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
190 size_t size)
191{
192 return snprintf(val, size, "%lld", ci->i_subdirs);
193}
194
195static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
196 size_t size)
197{
198 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
199}
200
201static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
202 size_t size)
203{
204 return snprintf(val, size, "%lld", ci->i_rfiles);
205}
206
207static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
208 size_t size)
209{
210 return snprintf(val, size, "%lld", ci->i_rsubdirs);
211}
212
213static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
214 size_t size)
215{
216 return snprintf(val, size, "%lld", ci->i_rbytes);
217}
218
219static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
220 size_t size)
221{
222 return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
223 (long)ci->i_rctime.tv_nsec);
224}
225
226
227#define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
228#define CEPH_XATTR_NAME2(_type, _name, _name2) \
229 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
230
231#define XATTR_NAME_CEPH(_type, _name) \
232 { \
233 .name = CEPH_XATTR_NAME(_type, _name), \
234 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
235 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
236 .readonly = true, \
237 .hidden = false, \
238 .exists_cb = NULL, \
239 }
240#define XATTR_LAYOUT_FIELD(_type, _name, _field) \
241 { \
242 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
243 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
244 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
245 .readonly = false, \
246 .hidden = true, \
247 .exists_cb = ceph_vxattrcb_layout_exists, \
248 }
249
250static struct ceph_vxattr ceph_dir_vxattrs[] = {
251 {
252 .name = "ceph.dir.layout",
253 .name_size = sizeof("ceph.dir.layout"),
254 .getxattr_cb = ceph_vxattrcb_layout,
255 .readonly = false,
256 .hidden = true,
257 .exists_cb = ceph_vxattrcb_layout_exists,
258 },
259 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
260 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
261 XATTR_LAYOUT_FIELD(dir, layout, object_size),
262 XATTR_LAYOUT_FIELD(dir, layout, pool),
263 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
264 XATTR_NAME_CEPH(dir, entries),
265 XATTR_NAME_CEPH(dir, files),
266 XATTR_NAME_CEPH(dir, subdirs),
267 XATTR_NAME_CEPH(dir, rentries),
268 XATTR_NAME_CEPH(dir, rfiles),
269 XATTR_NAME_CEPH(dir, rsubdirs),
270 XATTR_NAME_CEPH(dir, rbytes),
271 XATTR_NAME_CEPH(dir, rctime),
272 { .name = NULL, 0 } /* Required table terminator */
273};
274static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
275
276/* files */
277
278static struct ceph_vxattr ceph_file_vxattrs[] = {
279 {
280 .name = "ceph.file.layout",
281 .name_size = sizeof("ceph.file.layout"),
282 .getxattr_cb = ceph_vxattrcb_layout,
283 .readonly = false,
284 .hidden = true,
285 .exists_cb = ceph_vxattrcb_layout_exists,
286 },
287 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
288 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
289 XATTR_LAYOUT_FIELD(file, layout, object_size),
290 XATTR_LAYOUT_FIELD(file, layout, pool),
291 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
292 { .name = NULL, 0 } /* Required table terminator */
293};
294static size_t ceph_file_vxattrs_name_size; /* total size of all names */
295
296static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
297{
298 if (S_ISDIR(inode->i_mode))
299 return ceph_dir_vxattrs;
300 else if (S_ISREG(inode->i_mode))
301 return ceph_file_vxattrs;
302 return NULL;
303}
304
305static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
306{
307 if (vxattrs == ceph_dir_vxattrs)
308 return ceph_dir_vxattrs_name_size;
309 if (vxattrs == ceph_file_vxattrs)
310 return ceph_file_vxattrs_name_size;
311 BUG_ON(vxattrs);
312 return 0;
313}
314
315/*
316 * Compute the aggregate size (including terminating '\0') of all
317 * virtual extended attribute names in the given vxattr table.
318 */
319static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
320{
321 struct ceph_vxattr *vxattr;
322 size_t size = 0;
323
324 for (vxattr = vxattrs; vxattr->name; vxattr++)
325 if (!vxattr->hidden)
326 size += vxattr->name_size;
327
328 return size;
329}
330
331/* Routines called at initialization and exit time */
332
333void __init ceph_xattr_init(void)
334{
335 ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
336 ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
337}
338
339void ceph_xattr_exit(void)
340{
341 ceph_dir_vxattrs_name_size = 0;
342 ceph_file_vxattrs_name_size = 0;
343}
344
345static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
346 const char *name)
347{
348 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
349
350 if (vxattr) {
351 while (vxattr->name) {
352 if (!strcmp(vxattr->name, name))
353 return vxattr;
354 vxattr++;
355 }
356 }
357
358 return NULL;
359}
360
361static int __set_xattr(struct ceph_inode_info *ci,
362 const char *name, int name_len,
363 const char *val, int val_len,
364 int flags, int update_xattr,
365 struct ceph_inode_xattr **newxattr)
366{
367 struct rb_node **p;
368 struct rb_node *parent = NULL;
369 struct ceph_inode_xattr *xattr = NULL;
370 int c;
371 int new = 0;
372
373 p = &ci->i_xattrs.index.rb_node;
374 while (*p) {
375 parent = *p;
376 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
377 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
378 if (c < 0)
379 p = &(*p)->rb_left;
380 else if (c > 0)
381 p = &(*p)->rb_right;
382 else {
383 if (name_len == xattr->name_len)
384 break;
385 else if (name_len < xattr->name_len)
386 p = &(*p)->rb_left;
387 else
388 p = &(*p)->rb_right;
389 }
390 xattr = NULL;
391 }
392
393 if (update_xattr) {
394 int err = 0;
395 if (xattr && (flags & XATTR_CREATE))
396 err = -EEXIST;
397 else if (!xattr && (flags & XATTR_REPLACE))
398 err = -ENODATA;
399 if (err) {
400 kfree(name);
401 kfree(val);
402 return err;
403 }
404 if (update_xattr < 0) {
405 if (xattr)
406 __remove_xattr(ci, xattr);
407 kfree(name);
408 return 0;
409 }
410 }
411
412 if (!xattr) {
413 new = 1;
414 xattr = *newxattr;
415 xattr->name = name;
416 xattr->name_len = name_len;
417 xattr->should_free_name = update_xattr;
418
419 ci->i_xattrs.count++;
420 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
421 } else {
422 kfree(*newxattr);
423 *newxattr = NULL;
424 if (xattr->should_free_val)
425 kfree((void *)xattr->val);
426
427 if (update_xattr) {
428 kfree((void *)name);
429 name = xattr->name;
430 }
431 ci->i_xattrs.names_size -= xattr->name_len;
432 ci->i_xattrs.vals_size -= xattr->val_len;
433 }
434 ci->i_xattrs.names_size += name_len;
435 ci->i_xattrs.vals_size += val_len;
436 if (val)
437 xattr->val = val;
438 else
439 xattr->val = "";
440
441 xattr->val_len = val_len;
442 xattr->dirty = update_xattr;
443 xattr->should_free_val = (val && update_xattr);
444
445 if (new) {
446 rb_link_node(&xattr->node, parent, p);
447 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
448 dout("__set_xattr_val p=%p\n", p);
449 }
450
451 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
452 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
453
454 return 0;
455}
456
457static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
458 const char *name)
459{
460 struct rb_node **p;
461 struct rb_node *parent = NULL;
462 struct ceph_inode_xattr *xattr = NULL;
463 int name_len = strlen(name);
464 int c;
465
466 p = &ci->i_xattrs.index.rb_node;
467 while (*p) {
468 parent = *p;
469 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
470 c = strncmp(name, xattr->name, xattr->name_len);
471 if (c == 0 && name_len > xattr->name_len)
472 c = 1;
473 if (c < 0)
474 p = &(*p)->rb_left;
475 else if (c > 0)
476 p = &(*p)->rb_right;
477 else {
478 dout("__get_xattr %s: found %.*s\n", name,
479 xattr->val_len, xattr->val);
480 return xattr;
481 }
482 }
483
484 dout("__get_xattr %s: not found\n", name);
485
486 return NULL;
487}
488
489static void __free_xattr(struct ceph_inode_xattr *xattr)
490{
491 BUG_ON(!xattr);
492
493 if (xattr->should_free_name)
494 kfree((void *)xattr->name);
495 if (xattr->should_free_val)
496 kfree((void *)xattr->val);
497
498 kfree(xattr);
499}
500
501static int __remove_xattr(struct ceph_inode_info *ci,
502 struct ceph_inode_xattr *xattr)
503{
504 if (!xattr)
505 return -ENODATA;
506
507 rb_erase(&xattr->node, &ci->i_xattrs.index);
508
509 if (xattr->should_free_name)
510 kfree((void *)xattr->name);
511 if (xattr->should_free_val)
512 kfree((void *)xattr->val);
513
514 ci->i_xattrs.names_size -= xattr->name_len;
515 ci->i_xattrs.vals_size -= xattr->val_len;
516 ci->i_xattrs.count--;
517 kfree(xattr);
518
519 return 0;
520}
521
522static char *__copy_xattr_names(struct ceph_inode_info *ci,
523 char *dest)
524{
525 struct rb_node *p;
526 struct ceph_inode_xattr *xattr = NULL;
527
528 p = rb_first(&ci->i_xattrs.index);
529 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
530
531 while (p) {
532 xattr = rb_entry(p, struct ceph_inode_xattr, node);
533 memcpy(dest, xattr->name, xattr->name_len);
534 dest[xattr->name_len] = '\0';
535
536 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
537 xattr->name_len, ci->i_xattrs.names_size);
538
539 dest += xattr->name_len + 1;
540 p = rb_next(p);
541 }
542
543 return dest;
544}
545
546void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
547{
548 struct rb_node *p, *tmp;
549 struct ceph_inode_xattr *xattr = NULL;
550
551 p = rb_first(&ci->i_xattrs.index);
552
553 dout("__ceph_destroy_xattrs p=%p\n", p);
554
555 while (p) {
556 xattr = rb_entry(p, struct ceph_inode_xattr, node);
557 tmp = p;
558 p = rb_next(tmp);
559 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
560 xattr->name_len, xattr->name);
561 rb_erase(tmp, &ci->i_xattrs.index);
562
563 __free_xattr(xattr);
564 }
565
566 ci->i_xattrs.names_size = 0;
567 ci->i_xattrs.vals_size = 0;
568 ci->i_xattrs.index_version = 0;
569 ci->i_xattrs.count = 0;
570 ci->i_xattrs.index = RB_ROOT;
571}
572
573static int __build_xattrs(struct inode *inode)
574 __releases(ci->i_ceph_lock)
575 __acquires(ci->i_ceph_lock)
576{
577 u32 namelen;
578 u32 numattr = 0;
579 void *p, *end;
580 u32 len;
581 const char *name, *val;
582 struct ceph_inode_info *ci = ceph_inode(inode);
583 int xattr_version;
584 struct ceph_inode_xattr **xattrs = NULL;
585 int err = 0;
586 int i;
587
588 dout("__build_xattrs() len=%d\n",
589 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
590
591 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
592 return 0; /* already built */
593
594 __ceph_destroy_xattrs(ci);
595
596start:
597 /* updated internal xattr rb tree */
598 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
599 p = ci->i_xattrs.blob->vec.iov_base;
600 end = p + ci->i_xattrs.blob->vec.iov_len;
601 ceph_decode_32_safe(&p, end, numattr, bad);
602 xattr_version = ci->i_xattrs.version;
603 spin_unlock(&ci->i_ceph_lock);
604
605 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
606 GFP_NOFS);
607 err = -ENOMEM;
608 if (!xattrs)
609 goto bad_lock;
610
611 for (i = 0; i < numattr; i++) {
612 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
613 GFP_NOFS);
614 if (!xattrs[i])
615 goto bad_lock;
616 }
617
618 spin_lock(&ci->i_ceph_lock);
619 if (ci->i_xattrs.version != xattr_version) {
620 /* lost a race, retry */
621 for (i = 0; i < numattr; i++)
622 kfree(xattrs[i]);
623 kfree(xattrs);
624 xattrs = NULL;
625 goto start;
626 }
627 err = -EIO;
628 while (numattr--) {
629 ceph_decode_32_safe(&p, end, len, bad);
630 namelen = len;
631 name = p;
632 p += len;
633 ceph_decode_32_safe(&p, end, len, bad);
634 val = p;
635 p += len;
636
637 err = __set_xattr(ci, name, namelen, val, len,
638 0, 0, &xattrs[numattr]);
639
640 if (err < 0)
641 goto bad;
642 }
643 kfree(xattrs);
644 }
645 ci->i_xattrs.index_version = ci->i_xattrs.version;
646 ci->i_xattrs.dirty = false;
647
648 return err;
649bad_lock:
650 spin_lock(&ci->i_ceph_lock);
651bad:
652 if (xattrs) {
653 for (i = 0; i < numattr; i++)
654 kfree(xattrs[i]);
655 kfree(xattrs);
656 }
657 ci->i_xattrs.names_size = 0;
658 return err;
659}
660
661static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
662 int val_size)
663{
664 /*
665 * 4 bytes for the length, and additional 4 bytes per each xattr name,
666 * 4 bytes per each value
667 */
668 int size = 4 + ci->i_xattrs.count*(4 + 4) +
669 ci->i_xattrs.names_size +
670 ci->i_xattrs.vals_size;
671 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
672 ci->i_xattrs.count, ci->i_xattrs.names_size,
673 ci->i_xattrs.vals_size);
674
675 if (name_size)
676 size += 4 + 4 + name_size + val_size;
677
678 return size;
679}
680
681/*
682 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
683 * and swap into place.
684 */
685void __ceph_build_xattrs_blob(struct ceph_inode_info *ci)
686{
687 struct rb_node *p;
688 struct ceph_inode_xattr *xattr = NULL;
689 void *dest;
690
691 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
692 if (ci->i_xattrs.dirty) {
693 int need = __get_required_blob_size(ci, 0, 0);
694
695 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
696
697 p = rb_first(&ci->i_xattrs.index);
698 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
699
700 ceph_encode_32(&dest, ci->i_xattrs.count);
701 while (p) {
702 xattr = rb_entry(p, struct ceph_inode_xattr, node);
703
704 ceph_encode_32(&dest, xattr->name_len);
705 memcpy(dest, xattr->name, xattr->name_len);
706 dest += xattr->name_len;
707 ceph_encode_32(&dest, xattr->val_len);
708 memcpy(dest, xattr->val, xattr->val_len);
709 dest += xattr->val_len;
710
711 p = rb_next(p);
712 }
713
714 /* adjust buffer len; it may be larger than we need */
715 ci->i_xattrs.prealloc_blob->vec.iov_len =
716 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
717
718 if (ci->i_xattrs.blob)
719 ceph_buffer_put(ci->i_xattrs.blob);
720 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
721 ci->i_xattrs.prealloc_blob = NULL;
722 ci->i_xattrs.dirty = false;
723 ci->i_xattrs.version++;
724 }
725}
726
727static inline int __get_request_mask(struct inode *in) {
728 struct ceph_mds_request *req = current->journal_info;
729 int mask = 0;
730 if (req && req->r_target_inode == in) {
731 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
732 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
733 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
734 req->r_op == CEPH_MDS_OP_GETATTR) {
735 mask = le32_to_cpu(req->r_args.getattr.mask);
736 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
737 req->r_op == CEPH_MDS_OP_CREATE) {
738 mask = le32_to_cpu(req->r_args.open.mask);
739 }
740 }
741 return mask;
742}
743
744ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
745 size_t size)
746{
747 struct ceph_inode_info *ci = ceph_inode(inode);
748 struct ceph_inode_xattr *xattr;
749 struct ceph_vxattr *vxattr = NULL;
750 int req_mask;
751 int err;
752
753 /* let's see if a virtual xattr was requested */
754 vxattr = ceph_match_vxattr(inode, name);
755 if (vxattr) {
756 err = -ENODATA;
757 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci)))
758 err = vxattr->getxattr_cb(ci, value, size);
759 return err;
760 }
761
762 req_mask = __get_request_mask(inode);
763
764 spin_lock(&ci->i_ceph_lock);
765 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
766 ci->i_xattrs.version, ci->i_xattrs.index_version);
767
768 if (ci->i_xattrs.version == 0 ||
769 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
770 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
771 spin_unlock(&ci->i_ceph_lock);
772
773 /* security module gets xattr while filling trace */
774 if (current->journal_info != NULL) {
775 pr_warn_ratelimited("sync getxattr %p "
776 "during filling trace\n", inode);
777 return -EBUSY;
778 }
779
780 /* get xattrs from mds (if we don't already have them) */
781 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
782 if (err)
783 return err;
784 spin_lock(&ci->i_ceph_lock);
785 }
786
787 err = __build_xattrs(inode);
788 if (err < 0)
789 goto out;
790
791 err = -ENODATA; /* == ENOATTR */
792 xattr = __get_xattr(ci, name);
793 if (!xattr)
794 goto out;
795
796 err = -ERANGE;
797 if (size && size < xattr->val_len)
798 goto out;
799
800 err = xattr->val_len;
801 if (size == 0)
802 goto out;
803
804 memcpy(value, xattr->val, xattr->val_len);
805
806 if (current->journal_info != NULL &&
807 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
808 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
809out:
810 spin_unlock(&ci->i_ceph_lock);
811 return err;
812}
813
814ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
815{
816 struct inode *inode = d_inode(dentry);
817 struct ceph_inode_info *ci = ceph_inode(inode);
818 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
819 u32 vir_namelen = 0;
820 u32 namelen;
821 int err;
822 u32 len;
823 int i;
824
825 spin_lock(&ci->i_ceph_lock);
826 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
827 ci->i_xattrs.version, ci->i_xattrs.index_version);
828
829 if (ci->i_xattrs.version == 0 ||
830 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
831 spin_unlock(&ci->i_ceph_lock);
832 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
833 if (err)
834 return err;
835 spin_lock(&ci->i_ceph_lock);
836 }
837
838 err = __build_xattrs(inode);
839 if (err < 0)
840 goto out;
841 /*
842 * Start with virtual dir xattr names (if any) (including
843 * terminating '\0' characters for each).
844 */
845 vir_namelen = ceph_vxattrs_name_size(vxattrs);
846
847 /* adding 1 byte per each variable due to the null termination */
848 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
849 err = -ERANGE;
850 if (size && vir_namelen + namelen > size)
851 goto out;
852
853 err = namelen + vir_namelen;
854 if (size == 0)
855 goto out;
856
857 names = __copy_xattr_names(ci, names);
858
859 /* virtual xattr names, too */
860 err = namelen;
861 if (vxattrs) {
862 for (i = 0; vxattrs[i].name; i++) {
863 if (!vxattrs[i].hidden &&
864 !(vxattrs[i].exists_cb &&
865 !vxattrs[i].exists_cb(ci))) {
866 len = sprintf(names, "%s", vxattrs[i].name);
867 names += len + 1;
868 err += len + 1;
869 }
870 }
871 }
872
873out:
874 spin_unlock(&ci->i_ceph_lock);
875 return err;
876}
877
878static int ceph_sync_setxattr(struct inode *inode, const char *name,
879 const char *value, size_t size, int flags)
880{
881 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
882 struct ceph_inode_info *ci = ceph_inode(inode);
883 struct ceph_mds_request *req;
884 struct ceph_mds_client *mdsc = fsc->mdsc;
885 struct ceph_pagelist *pagelist = NULL;
886 int op = CEPH_MDS_OP_SETXATTR;
887 int err;
888
889 if (size > 0) {
890 /* copy value into pagelist */
891 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
892 if (!pagelist)
893 return -ENOMEM;
894
895 ceph_pagelist_init(pagelist);
896 err = ceph_pagelist_append(pagelist, value, size);
897 if (err)
898 goto out;
899 } else if (!value) {
900 if (flags & CEPH_XATTR_REPLACE)
901 op = CEPH_MDS_OP_RMXATTR;
902 else
903 flags |= CEPH_XATTR_REMOVE;
904 }
905
906 dout("setxattr value=%.*s\n", (int)size, value);
907
908 /* do request */
909 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
910 if (IS_ERR(req)) {
911 err = PTR_ERR(req);
912 goto out;
913 }
914
915 req->r_path2 = kstrdup(name, GFP_NOFS);
916 if (!req->r_path2) {
917 ceph_mdsc_put_request(req);
918 err = -ENOMEM;
919 goto out;
920 }
921
922 if (op == CEPH_MDS_OP_SETXATTR) {
923 req->r_args.setxattr.flags = cpu_to_le32(flags);
924 req->r_pagelist = pagelist;
925 pagelist = NULL;
926 }
927
928 req->r_inode = inode;
929 ihold(inode);
930 req->r_num_caps = 1;
931 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
932
933 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
934 err = ceph_mdsc_do_request(mdsc, NULL, req);
935 ceph_mdsc_put_request(req);
936 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
937
938out:
939 if (pagelist)
940 ceph_pagelist_release(pagelist);
941 return err;
942}
943
944int __ceph_setxattr(struct inode *inode, const char *name,
945 const void *value, size_t size, int flags)
946{
947 struct ceph_vxattr *vxattr;
948 struct ceph_inode_info *ci = ceph_inode(inode);
949 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
950 struct ceph_cap_flush *prealloc_cf = NULL;
951 int issued;
952 int err;
953 int dirty = 0;
954 int name_len = strlen(name);
955 int val_len = size;
956 char *newname = NULL;
957 char *newval = NULL;
958 struct ceph_inode_xattr *xattr = NULL;
959 int required_blob_size;
960 bool lock_snap_rwsem = false;
961
962 if (ceph_snap(inode) != CEPH_NOSNAP)
963 return -EROFS;
964
965 vxattr = ceph_match_vxattr(inode, name);
966 if (vxattr && vxattr->readonly)
967 return -EOPNOTSUPP;
968
969 /* pass any unhandled ceph.* xattrs through to the MDS */
970 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
971 goto do_sync_unlocked;
972
973 /* preallocate memory for xattr name, value, index node */
974 err = -ENOMEM;
975 newname = kmemdup(name, name_len + 1, GFP_NOFS);
976 if (!newname)
977 goto out;
978
979 if (val_len) {
980 newval = kmemdup(value, val_len, GFP_NOFS);
981 if (!newval)
982 goto out;
983 }
984
985 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
986 if (!xattr)
987 goto out;
988
989 prealloc_cf = ceph_alloc_cap_flush();
990 if (!prealloc_cf)
991 goto out;
992
993 spin_lock(&ci->i_ceph_lock);
994retry:
995 issued = __ceph_caps_issued(ci, NULL);
996 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
997 goto do_sync;
998
999 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1000 lock_snap_rwsem = true;
1001 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1002 spin_unlock(&ci->i_ceph_lock);
1003 down_read(&mdsc->snap_rwsem);
1004 spin_lock(&ci->i_ceph_lock);
1005 goto retry;
1006 }
1007 }
1008
1009 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1010 __build_xattrs(inode);
1011
1012 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1013
1014 if (!ci->i_xattrs.prealloc_blob ||
1015 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1016 struct ceph_buffer *blob;
1017
1018 spin_unlock(&ci->i_ceph_lock);
1019 dout(" preaallocating new blob size=%d\n", required_blob_size);
1020 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1021 if (!blob)
1022 goto do_sync_unlocked;
1023 spin_lock(&ci->i_ceph_lock);
1024 if (ci->i_xattrs.prealloc_blob)
1025 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
1026 ci->i_xattrs.prealloc_blob = blob;
1027 goto retry;
1028 }
1029
1030 err = __set_xattr(ci, newname, name_len, newval, val_len,
1031 flags, value ? 1 : -1, &xattr);
1032
1033 if (!err) {
1034 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1035 &prealloc_cf);
1036 ci->i_xattrs.dirty = true;
1037 inode->i_ctime = current_time(inode);
1038 }
1039
1040 spin_unlock(&ci->i_ceph_lock);
1041 if (lock_snap_rwsem)
1042 up_read(&mdsc->snap_rwsem);
1043 if (dirty)
1044 __mark_inode_dirty(inode, dirty);
1045 ceph_free_cap_flush(prealloc_cf);
1046 return err;
1047
1048do_sync:
1049 spin_unlock(&ci->i_ceph_lock);
1050do_sync_unlocked:
1051 if (lock_snap_rwsem)
1052 up_read(&mdsc->snap_rwsem);
1053
1054 /* security module set xattr while filling trace */
1055 if (current->journal_info != NULL) {
1056 pr_warn_ratelimited("sync setxattr %p "
1057 "during filling trace\n", inode);
1058 err = -EBUSY;
1059 } else {
1060 err = ceph_sync_setxattr(inode, name, value, size, flags);
1061 }
1062out:
1063 ceph_free_cap_flush(prealloc_cf);
1064 kfree(newname);
1065 kfree(newval);
1066 kfree(xattr);
1067 return err;
1068}
1069
1070static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1071 struct dentry *dentry, struct inode *inode,
1072 const char *name, void *value, size_t size)
1073{
1074 if (!ceph_is_valid_xattr(name))
1075 return -EOPNOTSUPP;
1076 return __ceph_getxattr(inode, name, value, size);
1077}
1078
1079static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1080 struct dentry *unused, struct inode *inode,
1081 const char *name, const void *value,
1082 size_t size, int flags)
1083{
1084 if (!ceph_is_valid_xattr(name))
1085 return -EOPNOTSUPP;
1086 return __ceph_setxattr(inode, name, value, size, flags);
1087}
1088
1089static const struct xattr_handler ceph_other_xattr_handler = {
1090 .prefix = "", /* match any name => handlers called with full name */
1091 .get = ceph_get_xattr_handler,
1092 .set = ceph_set_xattr_handler,
1093};
1094
1095#ifdef CONFIG_SECURITY
1096bool ceph_security_xattr_wanted(struct inode *in)
1097{
1098 return in->i_security != NULL;
1099}
1100
1101bool ceph_security_xattr_deadlock(struct inode *in)
1102{
1103 struct ceph_inode_info *ci;
1104 bool ret;
1105 if (in->i_security == NULL)
1106 return false;
1107 ci = ceph_inode(in);
1108 spin_lock(&ci->i_ceph_lock);
1109 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1110 !(ci->i_xattrs.version > 0 &&
1111 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1112 spin_unlock(&ci->i_ceph_lock);
1113 return ret;
1114}
1115#endif
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3#include <linux/ceph/pagelist.h>
4
5#include "super.h"
6#include "mds_client.h"
7
8#include <linux/ceph/decode.h>
9
10#include <linux/xattr.h>
11#include <linux/security.h>
12#include <linux/posix_acl_xattr.h>
13#include <linux/slab.h>
14
15#define XATTR_CEPH_PREFIX "ceph."
16#define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
17
18static int __remove_xattr(struct ceph_inode_info *ci,
19 struct ceph_inode_xattr *xattr);
20
21static bool ceph_is_valid_xattr(const char *name)
22{
23 return !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) ||
24 !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
25 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
26 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
27}
28
29/*
30 * These define virtual xattrs exposing the recursive directory
31 * statistics and layout metadata.
32 */
33struct ceph_vxattr {
34 char *name;
35 size_t name_size; /* strlen(name) + 1 (for '\0') */
36 ssize_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
37 size_t size);
38 bool (*exists_cb)(struct ceph_inode_info *ci);
39 unsigned int flags;
40};
41
42#define VXATTR_FLAG_READONLY (1<<0)
43#define VXATTR_FLAG_HIDDEN (1<<1)
44#define VXATTR_FLAG_RSTAT (1<<2)
45#define VXATTR_FLAG_DIRSTAT (1<<3)
46
47/* layouts */
48
49static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
50{
51 struct ceph_file_layout *fl = &ci->i_layout;
52 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
53 fl->object_size > 0 || fl->pool_id >= 0 ||
54 rcu_dereference_raw(fl->pool_ns) != NULL);
55}
56
57static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
58 size_t size)
59{
60 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
61 struct ceph_osd_client *osdc = &fsc->client->osdc;
62 struct ceph_string *pool_ns;
63 s64 pool = ci->i_layout.pool_id;
64 const char *pool_name;
65 const char *ns_field = " pool_namespace=";
66 char buf[128];
67 size_t len, total_len = 0;
68 ssize_t ret;
69
70 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
71
72 dout("ceph_vxattrcb_layout %p\n", &ci->netfs.inode);
73 down_read(&osdc->lock);
74 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
75 if (pool_name) {
76 len = snprintf(buf, sizeof(buf),
77 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
78 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
79 ci->i_layout.object_size);
80 total_len = len + strlen(pool_name);
81 } else {
82 len = snprintf(buf, sizeof(buf),
83 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
84 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
85 ci->i_layout.object_size, pool);
86 total_len = len;
87 }
88
89 if (pool_ns)
90 total_len += strlen(ns_field) + pool_ns->len;
91
92 ret = total_len;
93 if (size >= total_len) {
94 memcpy(val, buf, len);
95 ret = len;
96 if (pool_name) {
97 len = strlen(pool_name);
98 memcpy(val + ret, pool_name, len);
99 ret += len;
100 }
101 if (pool_ns) {
102 len = strlen(ns_field);
103 memcpy(val + ret, ns_field, len);
104 ret += len;
105 memcpy(val + ret, pool_ns->str, pool_ns->len);
106 ret += pool_ns->len;
107 }
108 }
109 up_read(&osdc->lock);
110 ceph_put_string(pool_ns);
111 return ret;
112}
113
114/*
115 * The convention with strings in xattrs is that they should not be NULL
116 * terminated, since we're returning the length with them. snprintf always
117 * NULL terminates however, so call it on a temporary buffer and then memcpy
118 * the result into place.
119 */
120static __printf(3, 4)
121int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
122{
123 int ret;
124 va_list args;
125 char buf[96]; /* NB: reevaluate size if new vxattrs are added */
126
127 va_start(args, fmt);
128 ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
129 va_end(args);
130
131 /* Sanity check */
132 if (size && ret + 1 > sizeof(buf)) {
133 WARN_ONCE(true, "Returned length too big (%d)", ret);
134 return -E2BIG;
135 }
136
137 if (ret <= size)
138 memcpy(val, buf, ret);
139 return ret;
140}
141
142static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
143 char *val, size_t size)
144{
145 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
146}
147
148static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
149 char *val, size_t size)
150{
151 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
152}
153
154static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
155 char *val, size_t size)
156{
157 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
158}
159
160static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
161 char *val, size_t size)
162{
163 ssize_t ret;
164 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
165 struct ceph_osd_client *osdc = &fsc->client->osdc;
166 s64 pool = ci->i_layout.pool_id;
167 const char *pool_name;
168
169 down_read(&osdc->lock);
170 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
171 if (pool_name) {
172 ret = strlen(pool_name);
173 if (ret <= size)
174 memcpy(val, pool_name, ret);
175 } else {
176 ret = ceph_fmt_xattr(val, size, "%lld", pool);
177 }
178 up_read(&osdc->lock);
179 return ret;
180}
181
182static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
183 char *val, size_t size)
184{
185 ssize_t ret = 0;
186 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
187
188 if (ns) {
189 ret = ns->len;
190 if (ret <= size)
191 memcpy(val, ns->str, ret);
192 ceph_put_string(ns);
193 }
194 return ret;
195}
196
197/* directories */
198
199static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
200 size_t size)
201{
202 return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
203}
204
205static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
206 size_t size)
207{
208 return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
209}
210
211static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
212 size_t size)
213{
214 return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
215}
216
217static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
218 size_t size)
219{
220 return ceph_fmt_xattr(val, size, "%lld",
221 ci->i_rfiles + ci->i_rsubdirs);
222}
223
224static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
225 size_t size)
226{
227 return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
228}
229
230static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
231 size_t size)
232{
233 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
234}
235
236static ssize_t ceph_vxattrcb_dir_rsnaps(struct ceph_inode_info *ci, char *val,
237 size_t size)
238{
239 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsnaps);
240}
241
242static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
243 size_t size)
244{
245 return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
246}
247
248static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
249 size_t size)
250{
251 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
252 ci->i_rctime.tv_nsec);
253}
254
255/* dir pin */
256static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
257{
258 return ci->i_dir_pin != -ENODATA;
259}
260
261static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
262 size_t size)
263{
264 return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
265}
266
267/* quotas */
268static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
269{
270 bool ret = false;
271 spin_lock(&ci->i_ceph_lock);
272 if ((ci->i_max_files || ci->i_max_bytes) &&
273 ci->i_vino.snap == CEPH_NOSNAP &&
274 ci->i_snap_realm &&
275 ci->i_snap_realm->ino == ci->i_vino.ino)
276 ret = true;
277 spin_unlock(&ci->i_ceph_lock);
278 return ret;
279}
280
281static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
282 size_t size)
283{
284 return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
285 ci->i_max_bytes, ci->i_max_files);
286}
287
288static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
289 char *val, size_t size)
290{
291 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
292}
293
294static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
295 char *val, size_t size)
296{
297 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
298}
299
300/* snapshots */
301static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
302{
303 return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
304}
305
306static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
307 size_t size)
308{
309 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
310 ci->i_snap_btime.tv_nsec);
311}
312
313static ssize_t ceph_vxattrcb_cluster_fsid(struct ceph_inode_info *ci,
314 char *val, size_t size)
315{
316 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
317
318 return ceph_fmt_xattr(val, size, "%pU", &fsc->client->fsid);
319}
320
321static ssize_t ceph_vxattrcb_client_id(struct ceph_inode_info *ci,
322 char *val, size_t size)
323{
324 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->netfs.inode.i_sb);
325
326 return ceph_fmt_xattr(val, size, "client%lld",
327 ceph_client_gid(fsc->client));
328}
329
330static ssize_t ceph_vxattrcb_caps(struct ceph_inode_info *ci, char *val,
331 size_t size)
332{
333 int issued;
334
335 spin_lock(&ci->i_ceph_lock);
336 issued = __ceph_caps_issued(ci, NULL);
337 spin_unlock(&ci->i_ceph_lock);
338
339 return ceph_fmt_xattr(val, size, "%s/0x%x",
340 ceph_cap_string(issued), issued);
341}
342
343static ssize_t ceph_vxattrcb_auth_mds(struct ceph_inode_info *ci,
344 char *val, size_t size)
345{
346 int ret;
347
348 spin_lock(&ci->i_ceph_lock);
349 ret = ceph_fmt_xattr(val, size, "%d",
350 ci->i_auth_cap ? ci->i_auth_cap->session->s_mds : -1);
351 spin_unlock(&ci->i_ceph_lock);
352 return ret;
353}
354
355#define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
356#define CEPH_XATTR_NAME2(_type, _name, _name2) \
357 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
358
359#define XATTR_NAME_CEPH(_type, _name, _flags) \
360 { \
361 .name = CEPH_XATTR_NAME(_type, _name), \
362 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
363 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
364 .exists_cb = NULL, \
365 .flags = (VXATTR_FLAG_READONLY | _flags), \
366 }
367#define XATTR_RSTAT_FIELD(_type, _name) \
368 XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
369#define XATTR_RSTAT_FIELD_UPDATABLE(_type, _name) \
370 { \
371 .name = CEPH_XATTR_NAME(_type, _name), \
372 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
373 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
374 .exists_cb = NULL, \
375 .flags = VXATTR_FLAG_RSTAT, \
376 }
377#define XATTR_LAYOUT_FIELD(_type, _name, _field) \
378 { \
379 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
380 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
381 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
382 .exists_cb = ceph_vxattrcb_layout_exists, \
383 .flags = VXATTR_FLAG_HIDDEN, \
384 }
385#define XATTR_QUOTA_FIELD(_type, _name) \
386 { \
387 .name = CEPH_XATTR_NAME(_type, _name), \
388 .name_size = sizeof(CEPH_XATTR_NAME(_type, _name)), \
389 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
390 .exists_cb = ceph_vxattrcb_quota_exists, \
391 .flags = VXATTR_FLAG_HIDDEN, \
392 }
393
394static struct ceph_vxattr ceph_dir_vxattrs[] = {
395 {
396 .name = "ceph.dir.layout",
397 .name_size = sizeof("ceph.dir.layout"),
398 .getxattr_cb = ceph_vxattrcb_layout,
399 .exists_cb = ceph_vxattrcb_layout_exists,
400 .flags = VXATTR_FLAG_HIDDEN,
401 },
402 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
403 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
404 XATTR_LAYOUT_FIELD(dir, layout, object_size),
405 XATTR_LAYOUT_FIELD(dir, layout, pool),
406 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
407 XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
408 XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
409 XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
410 XATTR_RSTAT_FIELD(dir, rentries),
411 XATTR_RSTAT_FIELD(dir, rfiles),
412 XATTR_RSTAT_FIELD(dir, rsubdirs),
413 XATTR_RSTAT_FIELD(dir, rsnaps),
414 XATTR_RSTAT_FIELD(dir, rbytes),
415 XATTR_RSTAT_FIELD_UPDATABLE(dir, rctime),
416 {
417 .name = "ceph.dir.pin",
418 .name_size = sizeof("ceph.dir.pin"),
419 .getxattr_cb = ceph_vxattrcb_dir_pin,
420 .exists_cb = ceph_vxattrcb_dir_pin_exists,
421 .flags = VXATTR_FLAG_HIDDEN,
422 },
423 {
424 .name = "ceph.quota",
425 .name_size = sizeof("ceph.quota"),
426 .getxattr_cb = ceph_vxattrcb_quota,
427 .exists_cb = ceph_vxattrcb_quota_exists,
428 .flags = VXATTR_FLAG_HIDDEN,
429 },
430 XATTR_QUOTA_FIELD(quota, max_bytes),
431 XATTR_QUOTA_FIELD(quota, max_files),
432 {
433 .name = "ceph.snap.btime",
434 .name_size = sizeof("ceph.snap.btime"),
435 .getxattr_cb = ceph_vxattrcb_snap_btime,
436 .exists_cb = ceph_vxattrcb_snap_btime_exists,
437 .flags = VXATTR_FLAG_READONLY,
438 },
439 {
440 .name = "ceph.caps",
441 .name_size = sizeof("ceph.caps"),
442 .getxattr_cb = ceph_vxattrcb_caps,
443 .exists_cb = NULL,
444 .flags = VXATTR_FLAG_HIDDEN,
445 },
446 { .name = NULL, 0 } /* Required table terminator */
447};
448
449/* files */
450
451static struct ceph_vxattr ceph_file_vxattrs[] = {
452 {
453 .name = "ceph.file.layout",
454 .name_size = sizeof("ceph.file.layout"),
455 .getxattr_cb = ceph_vxattrcb_layout,
456 .exists_cb = ceph_vxattrcb_layout_exists,
457 .flags = VXATTR_FLAG_HIDDEN,
458 },
459 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
460 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
461 XATTR_LAYOUT_FIELD(file, layout, object_size),
462 XATTR_LAYOUT_FIELD(file, layout, pool),
463 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
464 {
465 .name = "ceph.snap.btime",
466 .name_size = sizeof("ceph.snap.btime"),
467 .getxattr_cb = ceph_vxattrcb_snap_btime,
468 .exists_cb = ceph_vxattrcb_snap_btime_exists,
469 .flags = VXATTR_FLAG_READONLY,
470 },
471 {
472 .name = "ceph.caps",
473 .name_size = sizeof("ceph.caps"),
474 .getxattr_cb = ceph_vxattrcb_caps,
475 .exists_cb = NULL,
476 .flags = VXATTR_FLAG_HIDDEN,
477 },
478 { .name = NULL, 0 } /* Required table terminator */
479};
480
481static struct ceph_vxattr ceph_common_vxattrs[] = {
482 {
483 .name = "ceph.cluster_fsid",
484 .name_size = sizeof("ceph.cluster_fsid"),
485 .getxattr_cb = ceph_vxattrcb_cluster_fsid,
486 .exists_cb = NULL,
487 .flags = VXATTR_FLAG_READONLY,
488 },
489 {
490 .name = "ceph.client_id",
491 .name_size = sizeof("ceph.client_id"),
492 .getxattr_cb = ceph_vxattrcb_client_id,
493 .exists_cb = NULL,
494 .flags = VXATTR_FLAG_READONLY,
495 },
496 {
497 .name = "ceph.auth_mds",
498 .name_size = sizeof("ceph.auth_mds"),
499 .getxattr_cb = ceph_vxattrcb_auth_mds,
500 .exists_cb = NULL,
501 .flags = VXATTR_FLAG_READONLY,
502 },
503 { .name = NULL, 0 } /* Required table terminator */
504};
505
506static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
507{
508 if (S_ISDIR(inode->i_mode))
509 return ceph_dir_vxattrs;
510 else if (S_ISREG(inode->i_mode))
511 return ceph_file_vxattrs;
512 return NULL;
513}
514
515static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
516 const char *name)
517{
518 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
519
520 if (vxattr) {
521 while (vxattr->name) {
522 if (!strcmp(vxattr->name, name))
523 return vxattr;
524 vxattr++;
525 }
526 }
527
528 vxattr = ceph_common_vxattrs;
529 while (vxattr->name) {
530 if (!strcmp(vxattr->name, name))
531 return vxattr;
532 vxattr++;
533 }
534
535 return NULL;
536}
537
538static int __set_xattr(struct ceph_inode_info *ci,
539 const char *name, int name_len,
540 const char *val, int val_len,
541 int flags, int update_xattr,
542 struct ceph_inode_xattr **newxattr)
543{
544 struct rb_node **p;
545 struct rb_node *parent = NULL;
546 struct ceph_inode_xattr *xattr = NULL;
547 int c;
548 int new = 0;
549
550 p = &ci->i_xattrs.index.rb_node;
551 while (*p) {
552 parent = *p;
553 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
554 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
555 if (c < 0)
556 p = &(*p)->rb_left;
557 else if (c > 0)
558 p = &(*p)->rb_right;
559 else {
560 if (name_len == xattr->name_len)
561 break;
562 else if (name_len < xattr->name_len)
563 p = &(*p)->rb_left;
564 else
565 p = &(*p)->rb_right;
566 }
567 xattr = NULL;
568 }
569
570 if (update_xattr) {
571 int err = 0;
572
573 if (xattr && (flags & XATTR_CREATE))
574 err = -EEXIST;
575 else if (!xattr && (flags & XATTR_REPLACE))
576 err = -ENODATA;
577 if (err) {
578 kfree(name);
579 kfree(val);
580 kfree(*newxattr);
581 return err;
582 }
583 if (update_xattr < 0) {
584 if (xattr)
585 __remove_xattr(ci, xattr);
586 kfree(name);
587 kfree(*newxattr);
588 return 0;
589 }
590 }
591
592 if (!xattr) {
593 new = 1;
594 xattr = *newxattr;
595 xattr->name = name;
596 xattr->name_len = name_len;
597 xattr->should_free_name = update_xattr;
598
599 ci->i_xattrs.count++;
600 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
601 } else {
602 kfree(*newxattr);
603 *newxattr = NULL;
604 if (xattr->should_free_val)
605 kfree(xattr->val);
606
607 if (update_xattr) {
608 kfree(name);
609 name = xattr->name;
610 }
611 ci->i_xattrs.names_size -= xattr->name_len;
612 ci->i_xattrs.vals_size -= xattr->val_len;
613 }
614 ci->i_xattrs.names_size += name_len;
615 ci->i_xattrs.vals_size += val_len;
616 if (val)
617 xattr->val = val;
618 else
619 xattr->val = "";
620
621 xattr->val_len = val_len;
622 xattr->dirty = update_xattr;
623 xattr->should_free_val = (val && update_xattr);
624
625 if (new) {
626 rb_link_node(&xattr->node, parent, p);
627 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
628 dout("__set_xattr_val p=%p\n", p);
629 }
630
631 dout("__set_xattr_val added %llx.%llx xattr %p %.*s=%.*s\n",
632 ceph_vinop(&ci->netfs.inode), xattr, name_len, name, val_len, val);
633
634 return 0;
635}
636
637static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
638 const char *name)
639{
640 struct rb_node **p;
641 struct rb_node *parent = NULL;
642 struct ceph_inode_xattr *xattr = NULL;
643 int name_len = strlen(name);
644 int c;
645
646 p = &ci->i_xattrs.index.rb_node;
647 while (*p) {
648 parent = *p;
649 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
650 c = strncmp(name, xattr->name, xattr->name_len);
651 if (c == 0 && name_len > xattr->name_len)
652 c = 1;
653 if (c < 0)
654 p = &(*p)->rb_left;
655 else if (c > 0)
656 p = &(*p)->rb_right;
657 else {
658 dout("__get_xattr %s: found %.*s\n", name,
659 xattr->val_len, xattr->val);
660 return xattr;
661 }
662 }
663
664 dout("__get_xattr %s: not found\n", name);
665
666 return NULL;
667}
668
669static void __free_xattr(struct ceph_inode_xattr *xattr)
670{
671 BUG_ON(!xattr);
672
673 if (xattr->should_free_name)
674 kfree(xattr->name);
675 if (xattr->should_free_val)
676 kfree(xattr->val);
677
678 kfree(xattr);
679}
680
681static int __remove_xattr(struct ceph_inode_info *ci,
682 struct ceph_inode_xattr *xattr)
683{
684 if (!xattr)
685 return -ENODATA;
686
687 rb_erase(&xattr->node, &ci->i_xattrs.index);
688
689 if (xattr->should_free_name)
690 kfree(xattr->name);
691 if (xattr->should_free_val)
692 kfree(xattr->val);
693
694 ci->i_xattrs.names_size -= xattr->name_len;
695 ci->i_xattrs.vals_size -= xattr->val_len;
696 ci->i_xattrs.count--;
697 kfree(xattr);
698
699 return 0;
700}
701
702static char *__copy_xattr_names(struct ceph_inode_info *ci,
703 char *dest)
704{
705 struct rb_node *p;
706 struct ceph_inode_xattr *xattr = NULL;
707
708 p = rb_first(&ci->i_xattrs.index);
709 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
710
711 while (p) {
712 xattr = rb_entry(p, struct ceph_inode_xattr, node);
713 memcpy(dest, xattr->name, xattr->name_len);
714 dest[xattr->name_len] = '\0';
715
716 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
717 xattr->name_len, ci->i_xattrs.names_size);
718
719 dest += xattr->name_len + 1;
720 p = rb_next(p);
721 }
722
723 return dest;
724}
725
726void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
727{
728 struct rb_node *p, *tmp;
729 struct ceph_inode_xattr *xattr = NULL;
730
731 p = rb_first(&ci->i_xattrs.index);
732
733 dout("__ceph_destroy_xattrs p=%p\n", p);
734
735 while (p) {
736 xattr = rb_entry(p, struct ceph_inode_xattr, node);
737 tmp = p;
738 p = rb_next(tmp);
739 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
740 xattr->name_len, xattr->name);
741 rb_erase(tmp, &ci->i_xattrs.index);
742
743 __free_xattr(xattr);
744 }
745
746 ci->i_xattrs.names_size = 0;
747 ci->i_xattrs.vals_size = 0;
748 ci->i_xattrs.index_version = 0;
749 ci->i_xattrs.count = 0;
750 ci->i_xattrs.index = RB_ROOT;
751}
752
753static int __build_xattrs(struct inode *inode)
754 __releases(ci->i_ceph_lock)
755 __acquires(ci->i_ceph_lock)
756{
757 u32 namelen;
758 u32 numattr = 0;
759 void *p, *end;
760 u32 len;
761 const char *name, *val;
762 struct ceph_inode_info *ci = ceph_inode(inode);
763 u64 xattr_version;
764 struct ceph_inode_xattr **xattrs = NULL;
765 int err = 0;
766 int i;
767
768 dout("__build_xattrs() len=%d\n",
769 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
770
771 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
772 return 0; /* already built */
773
774 __ceph_destroy_xattrs(ci);
775
776start:
777 /* updated internal xattr rb tree */
778 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
779 p = ci->i_xattrs.blob->vec.iov_base;
780 end = p + ci->i_xattrs.blob->vec.iov_len;
781 ceph_decode_32_safe(&p, end, numattr, bad);
782 xattr_version = ci->i_xattrs.version;
783 spin_unlock(&ci->i_ceph_lock);
784
785 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
786 GFP_NOFS);
787 err = -ENOMEM;
788 if (!xattrs)
789 goto bad_lock;
790
791 for (i = 0; i < numattr; i++) {
792 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
793 GFP_NOFS);
794 if (!xattrs[i])
795 goto bad_lock;
796 }
797
798 spin_lock(&ci->i_ceph_lock);
799 if (ci->i_xattrs.version != xattr_version) {
800 /* lost a race, retry */
801 for (i = 0; i < numattr; i++)
802 kfree(xattrs[i]);
803 kfree(xattrs);
804 xattrs = NULL;
805 goto start;
806 }
807 err = -EIO;
808 while (numattr--) {
809 ceph_decode_32_safe(&p, end, len, bad);
810 namelen = len;
811 name = p;
812 p += len;
813 ceph_decode_32_safe(&p, end, len, bad);
814 val = p;
815 p += len;
816
817 err = __set_xattr(ci, name, namelen, val, len,
818 0, 0, &xattrs[numattr]);
819
820 if (err < 0)
821 goto bad;
822 }
823 kfree(xattrs);
824 }
825 ci->i_xattrs.index_version = ci->i_xattrs.version;
826 ci->i_xattrs.dirty = false;
827
828 return err;
829bad_lock:
830 spin_lock(&ci->i_ceph_lock);
831bad:
832 if (xattrs) {
833 for (i = 0; i < numattr; i++)
834 kfree(xattrs[i]);
835 kfree(xattrs);
836 }
837 ci->i_xattrs.names_size = 0;
838 return err;
839}
840
841static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
842 int val_size)
843{
844 /*
845 * 4 bytes for the length, and additional 4 bytes per each xattr name,
846 * 4 bytes per each value
847 */
848 int size = 4 + ci->i_xattrs.count*(4 + 4) +
849 ci->i_xattrs.names_size +
850 ci->i_xattrs.vals_size;
851 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
852 ci->i_xattrs.count, ci->i_xattrs.names_size,
853 ci->i_xattrs.vals_size);
854
855 if (name_size)
856 size += 4 + 4 + name_size + val_size;
857
858 return size;
859}
860
861/*
862 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
863 * and swap into place. It returns the old i_xattrs.blob (or NULL) so
864 * that it can be freed by the caller as the i_ceph_lock is likely to be
865 * held.
866 */
867struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
868{
869 struct rb_node *p;
870 struct ceph_inode_xattr *xattr = NULL;
871 struct ceph_buffer *old_blob = NULL;
872 void *dest;
873
874 dout("__build_xattrs_blob %p\n", &ci->netfs.inode);
875 if (ci->i_xattrs.dirty) {
876 int need = __get_required_blob_size(ci, 0, 0);
877
878 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
879
880 p = rb_first(&ci->i_xattrs.index);
881 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
882
883 ceph_encode_32(&dest, ci->i_xattrs.count);
884 while (p) {
885 xattr = rb_entry(p, struct ceph_inode_xattr, node);
886
887 ceph_encode_32(&dest, xattr->name_len);
888 memcpy(dest, xattr->name, xattr->name_len);
889 dest += xattr->name_len;
890 ceph_encode_32(&dest, xattr->val_len);
891 memcpy(dest, xattr->val, xattr->val_len);
892 dest += xattr->val_len;
893
894 p = rb_next(p);
895 }
896
897 /* adjust buffer len; it may be larger than we need */
898 ci->i_xattrs.prealloc_blob->vec.iov_len =
899 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
900
901 if (ci->i_xattrs.blob)
902 old_blob = ci->i_xattrs.blob;
903 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
904 ci->i_xattrs.prealloc_blob = NULL;
905 ci->i_xattrs.dirty = false;
906 ci->i_xattrs.version++;
907 }
908
909 return old_blob;
910}
911
912static inline int __get_request_mask(struct inode *in) {
913 struct ceph_mds_request *req = current->journal_info;
914 int mask = 0;
915 if (req && req->r_target_inode == in) {
916 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
917 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
918 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
919 req->r_op == CEPH_MDS_OP_GETATTR) {
920 mask = le32_to_cpu(req->r_args.getattr.mask);
921 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
922 req->r_op == CEPH_MDS_OP_CREATE) {
923 mask = le32_to_cpu(req->r_args.open.mask);
924 }
925 }
926 return mask;
927}
928
929ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
930 size_t size)
931{
932 struct ceph_inode_info *ci = ceph_inode(inode);
933 struct ceph_inode_xattr *xattr;
934 struct ceph_vxattr *vxattr;
935 int req_mask;
936 ssize_t err;
937
938 if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
939 goto handle_non_vxattrs;
940
941 /* let's see if a virtual xattr was requested */
942 vxattr = ceph_match_vxattr(inode, name);
943 if (vxattr) {
944 int mask = 0;
945 if (vxattr->flags & VXATTR_FLAG_RSTAT)
946 mask |= CEPH_STAT_RSTAT;
947 if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
948 mask |= CEPH_CAP_FILE_SHARED;
949 err = ceph_do_getattr(inode, mask, true);
950 if (err)
951 return err;
952 err = -ENODATA;
953 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
954 err = vxattr->getxattr_cb(ci, value, size);
955 if (size && size < err)
956 err = -ERANGE;
957 }
958 return err;
959 } else {
960 err = ceph_do_getvxattr(inode, name, value, size);
961 /* this would happen with a new client and old server combo */
962 if (err == -EOPNOTSUPP)
963 err = -ENODATA;
964 return err;
965 }
966handle_non_vxattrs:
967 req_mask = __get_request_mask(inode);
968
969 spin_lock(&ci->i_ceph_lock);
970 dout("getxattr %p name '%s' ver=%lld index_ver=%lld\n", inode, name,
971 ci->i_xattrs.version, ci->i_xattrs.index_version);
972
973 if (ci->i_xattrs.version == 0 ||
974 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
975 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
976 spin_unlock(&ci->i_ceph_lock);
977
978 /* security module gets xattr while filling trace */
979 if (current->journal_info) {
980 pr_warn_ratelimited("sync getxattr %p "
981 "during filling trace\n", inode);
982 return -EBUSY;
983 }
984
985 /* get xattrs from mds (if we don't already have them) */
986 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
987 if (err)
988 return err;
989 spin_lock(&ci->i_ceph_lock);
990 }
991
992 err = __build_xattrs(inode);
993 if (err < 0)
994 goto out;
995
996 err = -ENODATA; /* == ENOATTR */
997 xattr = __get_xattr(ci, name);
998 if (!xattr)
999 goto out;
1000
1001 err = -ERANGE;
1002 if (size && size < xattr->val_len)
1003 goto out;
1004
1005 err = xattr->val_len;
1006 if (size == 0)
1007 goto out;
1008
1009 memcpy(value, xattr->val, xattr->val_len);
1010
1011 if (current->journal_info &&
1012 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
1013 security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
1014 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
1015out:
1016 spin_unlock(&ci->i_ceph_lock);
1017 return err;
1018}
1019
1020ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1021{
1022 struct inode *inode = d_inode(dentry);
1023 struct ceph_inode_info *ci = ceph_inode(inode);
1024 bool len_only = (size == 0);
1025 u32 namelen;
1026 int err;
1027
1028 spin_lock(&ci->i_ceph_lock);
1029 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
1030 ci->i_xattrs.version, ci->i_xattrs.index_version);
1031
1032 if (ci->i_xattrs.version == 0 ||
1033 !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1034 spin_unlock(&ci->i_ceph_lock);
1035 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1036 if (err)
1037 return err;
1038 spin_lock(&ci->i_ceph_lock);
1039 }
1040
1041 err = __build_xattrs(inode);
1042 if (err < 0)
1043 goto out;
1044
1045 /* add 1 byte for each xattr due to the null termination */
1046 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1047 if (!len_only) {
1048 if (namelen > size) {
1049 err = -ERANGE;
1050 goto out;
1051 }
1052 names = __copy_xattr_names(ci, names);
1053 size -= namelen;
1054 }
1055 err = namelen;
1056out:
1057 spin_unlock(&ci->i_ceph_lock);
1058 return err;
1059}
1060
1061static int ceph_sync_setxattr(struct inode *inode, const char *name,
1062 const char *value, size_t size, int flags)
1063{
1064 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1065 struct ceph_inode_info *ci = ceph_inode(inode);
1066 struct ceph_mds_request *req;
1067 struct ceph_mds_client *mdsc = fsc->mdsc;
1068 struct ceph_osd_client *osdc = &fsc->client->osdc;
1069 struct ceph_pagelist *pagelist = NULL;
1070 int op = CEPH_MDS_OP_SETXATTR;
1071 int err;
1072
1073 if (size > 0) {
1074 /* copy value into pagelist */
1075 pagelist = ceph_pagelist_alloc(GFP_NOFS);
1076 if (!pagelist)
1077 return -ENOMEM;
1078
1079 err = ceph_pagelist_append(pagelist, value, size);
1080 if (err)
1081 goto out;
1082 } else if (!value) {
1083 if (flags & CEPH_XATTR_REPLACE)
1084 op = CEPH_MDS_OP_RMXATTR;
1085 else
1086 flags |= CEPH_XATTR_REMOVE;
1087 }
1088
1089 dout("setxattr value size: %zu\n", size);
1090
1091 /* do request */
1092 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1093 if (IS_ERR(req)) {
1094 err = PTR_ERR(req);
1095 goto out;
1096 }
1097
1098 req->r_path2 = kstrdup(name, GFP_NOFS);
1099 if (!req->r_path2) {
1100 ceph_mdsc_put_request(req);
1101 err = -ENOMEM;
1102 goto out;
1103 }
1104
1105 if (op == CEPH_MDS_OP_SETXATTR) {
1106 req->r_args.setxattr.flags = cpu_to_le32(flags);
1107 req->r_args.setxattr.osdmap_epoch =
1108 cpu_to_le32(osdc->osdmap->epoch);
1109 req->r_pagelist = pagelist;
1110 pagelist = NULL;
1111 }
1112
1113 req->r_inode = inode;
1114 ihold(inode);
1115 req->r_num_caps = 1;
1116 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1117
1118 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1119 err = ceph_mdsc_do_request(mdsc, NULL, req);
1120 ceph_mdsc_put_request(req);
1121 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1122
1123out:
1124 if (pagelist)
1125 ceph_pagelist_release(pagelist);
1126 return err;
1127}
1128
1129int __ceph_setxattr(struct inode *inode, const char *name,
1130 const void *value, size_t size, int flags)
1131{
1132 struct ceph_vxattr *vxattr;
1133 struct ceph_inode_info *ci = ceph_inode(inode);
1134 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1135 struct ceph_cap_flush *prealloc_cf = NULL;
1136 struct ceph_buffer *old_blob = NULL;
1137 int issued;
1138 int err;
1139 int dirty = 0;
1140 int name_len = strlen(name);
1141 int val_len = size;
1142 char *newname = NULL;
1143 char *newval = NULL;
1144 struct ceph_inode_xattr *xattr = NULL;
1145 int required_blob_size;
1146 bool check_realm = false;
1147 bool lock_snap_rwsem = false;
1148
1149 if (ceph_snap(inode) != CEPH_NOSNAP)
1150 return -EROFS;
1151
1152 vxattr = ceph_match_vxattr(inode, name);
1153 if (vxattr) {
1154 if (vxattr->flags & VXATTR_FLAG_READONLY)
1155 return -EOPNOTSUPP;
1156 if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1157 check_realm = true;
1158 }
1159
1160 /* pass any unhandled ceph.* xattrs through to the MDS */
1161 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1162 goto do_sync_unlocked;
1163
1164 /* preallocate memory for xattr name, value, index node */
1165 err = -ENOMEM;
1166 newname = kmemdup(name, name_len + 1, GFP_NOFS);
1167 if (!newname)
1168 goto out;
1169
1170 if (val_len) {
1171 newval = kmemdup(value, val_len, GFP_NOFS);
1172 if (!newval)
1173 goto out;
1174 }
1175
1176 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1177 if (!xattr)
1178 goto out;
1179
1180 prealloc_cf = ceph_alloc_cap_flush();
1181 if (!prealloc_cf)
1182 goto out;
1183
1184 spin_lock(&ci->i_ceph_lock);
1185retry:
1186 issued = __ceph_caps_issued(ci, NULL);
1187 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1188 if ((ci->i_xattrs.version == 0) || !(issued & CEPH_CAP_XATTR_EXCL) ||
1189 (required_blob_size > mdsc->mdsmap->m_max_xattr_size)) {
1190 dout("%s do sync setxattr: version: %llu size: %d max: %llu\n",
1191 __func__, ci->i_xattrs.version, required_blob_size,
1192 mdsc->mdsmap->m_max_xattr_size);
1193 goto do_sync;
1194 }
1195
1196 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1197 lock_snap_rwsem = true;
1198 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1199 spin_unlock(&ci->i_ceph_lock);
1200 down_read(&mdsc->snap_rwsem);
1201 spin_lock(&ci->i_ceph_lock);
1202 goto retry;
1203 }
1204 }
1205
1206 dout("setxattr %p name '%s' issued %s\n", inode, name,
1207 ceph_cap_string(issued));
1208 __build_xattrs(inode);
1209
1210 if (!ci->i_xattrs.prealloc_blob ||
1211 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1212 struct ceph_buffer *blob;
1213
1214 spin_unlock(&ci->i_ceph_lock);
1215 ceph_buffer_put(old_blob); /* Shouldn't be required */
1216 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1217 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1218 if (!blob)
1219 goto do_sync_unlocked;
1220 spin_lock(&ci->i_ceph_lock);
1221 /* prealloc_blob can't be released while holding i_ceph_lock */
1222 if (ci->i_xattrs.prealloc_blob)
1223 old_blob = ci->i_xattrs.prealloc_blob;
1224 ci->i_xattrs.prealloc_blob = blob;
1225 goto retry;
1226 }
1227
1228 err = __set_xattr(ci, newname, name_len, newval, val_len,
1229 flags, value ? 1 : -1, &xattr);
1230
1231 if (!err) {
1232 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1233 &prealloc_cf);
1234 ci->i_xattrs.dirty = true;
1235 inode->i_ctime = current_time(inode);
1236 }
1237
1238 spin_unlock(&ci->i_ceph_lock);
1239 ceph_buffer_put(old_blob);
1240 if (lock_snap_rwsem)
1241 up_read(&mdsc->snap_rwsem);
1242 if (dirty)
1243 __mark_inode_dirty(inode, dirty);
1244 ceph_free_cap_flush(prealloc_cf);
1245 return err;
1246
1247do_sync:
1248 spin_unlock(&ci->i_ceph_lock);
1249do_sync_unlocked:
1250 if (lock_snap_rwsem)
1251 up_read(&mdsc->snap_rwsem);
1252
1253 /* security module set xattr while filling trace */
1254 if (current->journal_info) {
1255 pr_warn_ratelimited("sync setxattr %p "
1256 "during filling trace\n", inode);
1257 err = -EBUSY;
1258 } else {
1259 err = ceph_sync_setxattr(inode, name, value, size, flags);
1260 if (err >= 0 && check_realm) {
1261 /* check if snaprealm was created for quota inode */
1262 spin_lock(&ci->i_ceph_lock);
1263 if ((ci->i_max_files || ci->i_max_bytes) &&
1264 !(ci->i_snap_realm &&
1265 ci->i_snap_realm->ino == ci->i_vino.ino))
1266 err = -EOPNOTSUPP;
1267 spin_unlock(&ci->i_ceph_lock);
1268 }
1269 }
1270out:
1271 ceph_free_cap_flush(prealloc_cf);
1272 kfree(newname);
1273 kfree(newval);
1274 kfree(xattr);
1275 return err;
1276}
1277
1278static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1279 struct dentry *dentry, struct inode *inode,
1280 const char *name, void *value, size_t size)
1281{
1282 if (!ceph_is_valid_xattr(name))
1283 return -EOPNOTSUPP;
1284 return __ceph_getxattr(inode, name, value, size);
1285}
1286
1287static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1288 struct user_namespace *mnt_userns,
1289 struct dentry *unused, struct inode *inode,
1290 const char *name, const void *value,
1291 size_t size, int flags)
1292{
1293 if (!ceph_is_valid_xattr(name))
1294 return -EOPNOTSUPP;
1295 return __ceph_setxattr(inode, name, value, size, flags);
1296}
1297
1298static const struct xattr_handler ceph_other_xattr_handler = {
1299 .prefix = "", /* match any name => handlers called with full name */
1300 .get = ceph_get_xattr_handler,
1301 .set = ceph_set_xattr_handler,
1302};
1303
1304#ifdef CONFIG_SECURITY
1305bool ceph_security_xattr_wanted(struct inode *in)
1306{
1307 return in->i_security != NULL;
1308}
1309
1310bool ceph_security_xattr_deadlock(struct inode *in)
1311{
1312 struct ceph_inode_info *ci;
1313 bool ret;
1314 if (!in->i_security)
1315 return false;
1316 ci = ceph_inode(in);
1317 spin_lock(&ci->i_ceph_lock);
1318 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1319 !(ci->i_xattrs.version > 0 &&
1320 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1321 spin_unlock(&ci->i_ceph_lock);
1322 return ret;
1323}
1324
1325#ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1326int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1327 struct ceph_acl_sec_ctx *as_ctx)
1328{
1329 struct ceph_pagelist *pagelist = as_ctx->pagelist;
1330 const char *name;
1331 size_t name_len;
1332 int err;
1333
1334 err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1335 &name, &as_ctx->sec_ctx,
1336 &as_ctx->sec_ctxlen);
1337 if (err < 0) {
1338 WARN_ON_ONCE(err != -EOPNOTSUPP);
1339 err = 0; /* do nothing */
1340 goto out;
1341 }
1342
1343 err = -ENOMEM;
1344 if (!pagelist) {
1345 pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1346 if (!pagelist)
1347 goto out;
1348 err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1349 if (err)
1350 goto out;
1351 ceph_pagelist_encode_32(pagelist, 1);
1352 }
1353
1354 /*
1355 * FIXME: Make security_dentry_init_security() generic. Currently
1356 * It only supports single security module and only selinux has
1357 * dentry_init_security hook.
1358 */
1359 name_len = strlen(name);
1360 err = ceph_pagelist_reserve(pagelist,
1361 4 * 2 + name_len + as_ctx->sec_ctxlen);
1362 if (err)
1363 goto out;
1364
1365 if (as_ctx->pagelist) {
1366 /* update count of KV pairs */
1367 BUG_ON(pagelist->length <= sizeof(__le32));
1368 if (list_is_singular(&pagelist->head)) {
1369 le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1370 } else {
1371 struct page *page = list_first_entry(&pagelist->head,
1372 struct page, lru);
1373 void *addr = kmap_atomic(page);
1374 le32_add_cpu((__le32*)addr, 1);
1375 kunmap_atomic(addr);
1376 }
1377 } else {
1378 as_ctx->pagelist = pagelist;
1379 }
1380
1381 ceph_pagelist_encode_32(pagelist, name_len);
1382 ceph_pagelist_append(pagelist, name, name_len);
1383
1384 ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1385 ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1386
1387 err = 0;
1388out:
1389 if (pagelist && !as_ctx->pagelist)
1390 ceph_pagelist_release(pagelist);
1391 return err;
1392}
1393#endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1394#endif /* CONFIG_SECURITY */
1395
1396void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1397{
1398#ifdef CONFIG_CEPH_FS_POSIX_ACL
1399 posix_acl_release(as_ctx->acl);
1400 posix_acl_release(as_ctx->default_acl);
1401#endif
1402#ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1403 security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1404#endif
1405 if (as_ctx->pagelist)
1406 ceph_pagelist_release(as_ctx->pagelist);
1407}
1408
1409/*
1410 * List of handlers for synthetic system.* attributes. Other
1411 * attributes are handled directly.
1412 */
1413const struct xattr_handler *ceph_xattr_handlers[] = {
1414#ifdef CONFIG_CEPH_FS_POSIX_ACL
1415 &posix_acl_access_xattr_handler,
1416 &posix_acl_default_xattr_handler,
1417#endif
1418 &ceph_other_xattr_handler,
1419 NULL,
1420};