blob: 3a166f860b6cf7444785c64e60be8d43655eb8b6 [file] [log] [blame]
rjw1f884582022-01-06 17:20:42 +08001// 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/posix_acl_xattr.h>
12#include <linux/slab.h>
13
14#define XATTR_CEPH_PREFIX "ceph."
15#define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
16
17static int __remove_xattr(struct ceph_inode_info *ci,
18 struct ceph_inode_xattr *xattr);
19
20static const struct xattr_handler ceph_other_xattr_handler;
21
22/*
23 * List of handlers for synthetic system.* attributes. Other
24 * attributes are handled directly.
25 */
26const struct xattr_handler *ceph_xattr_handlers[] = {
27#ifdef CONFIG_CEPH_FS_POSIX_ACL
28 &posix_acl_access_xattr_handler,
29 &posix_acl_default_xattr_handler,
30#endif
31 &ceph_other_xattr_handler,
32 NULL,
33};
34
35static bool ceph_is_valid_xattr(const char *name)
36{
37 return !strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN) ||
38 !strncmp(name, XATTR_SECURITY_PREFIX,
39 XATTR_SECURITY_PREFIX_LEN) ||
40 !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
41 !strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
42}
43
44/*
45 * These define virtual xattrs exposing the recursive directory
46 * statistics and layout metadata.
47 */
48struct ceph_vxattr {
49 char *name;
50 size_t name_size; /* strlen(name) + 1 (for '\0') */
51 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
52 size_t size);
53 bool readonly, hidden;
54 bool (*exists_cb)(struct ceph_inode_info *ci);
55};
56
57/* layouts */
58
59static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
60{
61 struct ceph_file_layout *fl = &ci->i_layout;
62 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
63 fl->object_size > 0 || fl->pool_id >= 0 ||
64 rcu_dereference_raw(fl->pool_ns) != NULL);
65}
66
67static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
68 size_t size)
69{
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 struct ceph_string *pool_ns;
73 s64 pool = ci->i_layout.pool_id;
74 const char *pool_name;
75 const char *ns_field = " pool_namespace=";
76 char buf[128];
77 size_t len, total_len = 0;
78 ssize_t ret;
79
80 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
81
82 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
83 down_read(&osdc->lock);
84 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
85 if (pool_name) {
86 len = snprintf(buf, sizeof(buf),
87 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
88 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
89 ci->i_layout.object_size);
90 total_len = len + strlen(pool_name);
91 } else {
92 len = snprintf(buf, sizeof(buf),
93 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
94 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
95 ci->i_layout.object_size, (unsigned long long)pool);
96 total_len = len;
97 }
98
99 if (pool_ns)
100 total_len += strlen(ns_field) + pool_ns->len;
101
102 ret = total_len;
103 if (size >= total_len) {
104 memcpy(val, buf, len);
105 ret = len;
106 if (pool_name) {
107 len = strlen(pool_name);
108 memcpy(val + ret, pool_name, len);
109 ret += len;
110 }
111 if (pool_ns) {
112 len = strlen(ns_field);
113 memcpy(val + ret, ns_field, len);
114 ret += len;
115 memcpy(val + ret, pool_ns->str, pool_ns->len);
116 ret += pool_ns->len;
117 }
118 }
119 up_read(&osdc->lock);
120 ceph_put_string(pool_ns);
121 return ret;
122}
123
124static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
125 char *val, size_t size)
126{
127 return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
128}
129
130static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
131 char *val, size_t size)
132{
133 return snprintf(val, size, "%u", ci->i_layout.stripe_count);
134}
135
136static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
137 char *val, size_t size)
138{
139 return snprintf(val, size, "%u", ci->i_layout.object_size);
140}
141
142static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
143 char *val, size_t size)
144{
145 int ret;
146 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
147 struct ceph_osd_client *osdc = &fsc->client->osdc;
148 s64 pool = ci->i_layout.pool_id;
149 const char *pool_name;
150
151 down_read(&osdc->lock);
152 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
153 if (pool_name)
154 ret = snprintf(val, size, "%s", pool_name);
155 else
156 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
157 up_read(&osdc->lock);
158 return ret;
159}
160
161static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
162 char *val, size_t size)
163{
164 int ret = 0;
165 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
166 if (ns) {
167 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
168 ceph_put_string(ns);
169 }
170 return ret;
171}
172
173/* directories */
174
175static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
176 size_t size)
177{
178 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
179}
180
181static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
182 size_t size)
183{
184 return snprintf(val, size, "%lld", ci->i_files);
185}
186
187static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
188 size_t size)
189{
190 return snprintf(val, size, "%lld", ci->i_subdirs);
191}
192
193static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
194 size_t size)
195{
196 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
197}
198
199static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
200 size_t size)
201{
202 return snprintf(val, size, "%lld", ci->i_rfiles);
203}
204
205static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
206 size_t size)
207{
208 return snprintf(val, size, "%lld", ci->i_rsubdirs);
209}
210
211static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
212 size_t size)
213{
214 return snprintf(val, size, "%lld", ci->i_rbytes);
215}
216
217static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
218 size_t size)
219{
220 return snprintf(val, size, "%ld.09%ld", (long)ci->i_rctime.tv_sec,
221 (long)ci->i_rctime.tv_nsec);
222}
223
224
225#define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
226#define CEPH_XATTR_NAME2(_type, _name, _name2) \
227 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
228
229#define XATTR_NAME_CEPH(_type, _name) \
230 { \
231 .name = CEPH_XATTR_NAME(_type, _name), \
232 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
233 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
234 .readonly = true, \
235 .hidden = false, \
236 .exists_cb = NULL, \
237 }
238#define XATTR_LAYOUT_FIELD(_type, _name, _field) \
239 { \
240 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
241 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
242 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
243 .readonly = false, \
244 .hidden = true, \
245 .exists_cb = ceph_vxattrcb_layout_exists, \
246 }
247
248static struct ceph_vxattr ceph_dir_vxattrs[] = {
249 {
250 .name = "ceph.dir.layout",
251 .name_size = sizeof("ceph.dir.layout"),
252 .getxattr_cb = ceph_vxattrcb_layout,
253 .readonly = false,
254 .hidden = true,
255 .exists_cb = ceph_vxattrcb_layout_exists,
256 },
257 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
258 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
259 XATTR_LAYOUT_FIELD(dir, layout, object_size),
260 XATTR_LAYOUT_FIELD(dir, layout, pool),
261 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
262 XATTR_NAME_CEPH(dir, entries),
263 XATTR_NAME_CEPH(dir, files),
264 XATTR_NAME_CEPH(dir, subdirs),
265 XATTR_NAME_CEPH(dir, rentries),
266 XATTR_NAME_CEPH(dir, rfiles),
267 XATTR_NAME_CEPH(dir, rsubdirs),
268 XATTR_NAME_CEPH(dir, rbytes),
269 XATTR_NAME_CEPH(dir, rctime),
270 { .name = NULL, 0 } /* Required table terminator */
271};
272static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
273
274/* files */
275
276static struct ceph_vxattr ceph_file_vxattrs[] = {
277 {
278 .name = "ceph.file.layout",
279 .name_size = sizeof("ceph.file.layout"),
280 .getxattr_cb = ceph_vxattrcb_layout,
281 .readonly = false,
282 .hidden = true,
283 .exists_cb = ceph_vxattrcb_layout_exists,
284 },
285 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
286 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
287 XATTR_LAYOUT_FIELD(file, layout, object_size),
288 XATTR_LAYOUT_FIELD(file, layout, pool),
289 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
290 { .name = NULL, 0 } /* Required table terminator */
291};
292static size_t ceph_file_vxattrs_name_size; /* total size of all names */
293
294static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
295{
296 if (S_ISDIR(inode->i_mode))
297 return ceph_dir_vxattrs;
298 else if (S_ISREG(inode->i_mode))
299 return ceph_file_vxattrs;
300 return NULL;
301}
302
303static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
304{
305 if (vxattrs == ceph_dir_vxattrs)
306 return ceph_dir_vxattrs_name_size;
307 if (vxattrs == ceph_file_vxattrs)
308 return ceph_file_vxattrs_name_size;
309 BUG_ON(vxattrs);
310 return 0;
311}
312
313/*
314 * Compute the aggregate size (including terminating '\0') of all
315 * virtual extended attribute names in the given vxattr table.
316 */
317static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
318{
319 struct ceph_vxattr *vxattr;
320 size_t size = 0;
321
322 for (vxattr = vxattrs; vxattr->name; vxattr++)
323 if (!vxattr->hidden)
324 size += vxattr->name_size;
325
326 return size;
327}
328
329/* Routines called at initialization and exit time */
330
331void __init ceph_xattr_init(void)
332{
333 ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
334 ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
335}
336
337void ceph_xattr_exit(void)
338{
339 ceph_dir_vxattrs_name_size = 0;
340 ceph_file_vxattrs_name_size = 0;
341}
342
343static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
344 const char *name)
345{
346 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
347
348 if (vxattr) {
349 while (vxattr->name) {
350 if (!strcmp(vxattr->name, name))
351 return vxattr;
352 vxattr++;
353 }
354 }
355
356 return NULL;
357}
358
359static int __set_xattr(struct ceph_inode_info *ci,
360 const char *name, int name_len,
361 const char *val, int val_len,
362 int flags, int update_xattr,
363 struct ceph_inode_xattr **newxattr)
364{
365 struct rb_node **p;
366 struct rb_node *parent = NULL;
367 struct ceph_inode_xattr *xattr = NULL;
368 int c;
369 int new = 0;
370
371 p = &ci->i_xattrs.index.rb_node;
372 while (*p) {
373 parent = *p;
374 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
375 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
376 if (c < 0)
377 p = &(*p)->rb_left;
378 else if (c > 0)
379 p = &(*p)->rb_right;
380 else {
381 if (name_len == xattr->name_len)
382 break;
383 else if (name_len < xattr->name_len)
384 p = &(*p)->rb_left;
385 else
386 p = &(*p)->rb_right;
387 }
388 xattr = NULL;
389 }
390
391 if (update_xattr) {
392 int err = 0;
393
394 if (xattr && (flags & XATTR_CREATE))
395 err = -EEXIST;
396 else if (!xattr && (flags & XATTR_REPLACE))
397 err = -ENODATA;
398 if (err) {
399 kfree(name);
400 kfree(val);
401 kfree(*newxattr);
402 return err;
403 }
404 if (update_xattr < 0) {
405 if (xattr)
406 __remove_xattr(ci, xattr);
407 kfree(name);
408 kfree(*newxattr);
409 return 0;
410 }
411 }
412
413 if (!xattr) {
414 new = 1;
415 xattr = *newxattr;
416 xattr->name = name;
417 xattr->name_len = name_len;
418 xattr->should_free_name = update_xattr;
419
420 ci->i_xattrs.count++;
421 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
422 } else {
423 kfree(*newxattr);
424 *newxattr = NULL;
425 if (xattr->should_free_val)
426 kfree((void *)xattr->val);
427
428 if (update_xattr) {
429 kfree((void *)name);
430 name = xattr->name;
431 }
432 ci->i_xattrs.names_size -= xattr->name_len;
433 ci->i_xattrs.vals_size -= xattr->val_len;
434 }
435 ci->i_xattrs.names_size += name_len;
436 ci->i_xattrs.vals_size += val_len;
437 if (val)
438 xattr->val = val;
439 else
440 xattr->val = "";
441
442 xattr->val_len = val_len;
443 xattr->dirty = update_xattr;
444 xattr->should_free_val = (val && update_xattr);
445
446 if (new) {
447 rb_link_node(&xattr->node, parent, p);
448 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
449 dout("__set_xattr_val p=%p\n", p);
450 }
451
452 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
453 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
454
455 return 0;
456}
457
458static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
459 const char *name)
460{
461 struct rb_node **p;
462 struct rb_node *parent = NULL;
463 struct ceph_inode_xattr *xattr = NULL;
464 int name_len = strlen(name);
465 int c;
466
467 p = &ci->i_xattrs.index.rb_node;
468 while (*p) {
469 parent = *p;
470 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
471 c = strncmp(name, xattr->name, xattr->name_len);
472 if (c == 0 && name_len > xattr->name_len)
473 c = 1;
474 if (c < 0)
475 p = &(*p)->rb_left;
476 else if (c > 0)
477 p = &(*p)->rb_right;
478 else {
479 dout("__get_xattr %s: found %.*s\n", name,
480 xattr->val_len, xattr->val);
481 return xattr;
482 }
483 }
484
485 dout("__get_xattr %s: not found\n", name);
486
487 return NULL;
488}
489
490static void __free_xattr(struct ceph_inode_xattr *xattr)
491{
492 BUG_ON(!xattr);
493
494 if (xattr->should_free_name)
495 kfree((void *)xattr->name);
496 if (xattr->should_free_val)
497 kfree((void *)xattr->val);
498
499 kfree(xattr);
500}
501
502static int __remove_xattr(struct ceph_inode_info *ci,
503 struct ceph_inode_xattr *xattr)
504{
505 if (!xattr)
506 return -ENODATA;
507
508 rb_erase(&xattr->node, &ci->i_xattrs.index);
509
510 if (xattr->should_free_name)
511 kfree((void *)xattr->name);
512 if (xattr->should_free_val)
513 kfree((void *)xattr->val);
514
515 ci->i_xattrs.names_size -= xattr->name_len;
516 ci->i_xattrs.vals_size -= xattr->val_len;
517 ci->i_xattrs.count--;
518 kfree(xattr);
519
520 return 0;
521}
522
523static char *__copy_xattr_names(struct ceph_inode_info *ci,
524 char *dest)
525{
526 struct rb_node *p;
527 struct ceph_inode_xattr *xattr = NULL;
528
529 p = rb_first(&ci->i_xattrs.index);
530 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
531
532 while (p) {
533 xattr = rb_entry(p, struct ceph_inode_xattr, node);
534 memcpy(dest, xattr->name, xattr->name_len);
535 dest[xattr->name_len] = '\0';
536
537 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
538 xattr->name_len, ci->i_xattrs.names_size);
539
540 dest += xattr->name_len + 1;
541 p = rb_next(p);
542 }
543
544 return dest;
545}
546
547void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
548{
549 struct rb_node *p, *tmp;
550 struct ceph_inode_xattr *xattr = NULL;
551
552 p = rb_first(&ci->i_xattrs.index);
553
554 dout("__ceph_destroy_xattrs p=%p\n", p);
555
556 while (p) {
557 xattr = rb_entry(p, struct ceph_inode_xattr, node);
558 tmp = p;
559 p = rb_next(tmp);
560 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
561 xattr->name_len, xattr->name);
562 rb_erase(tmp, &ci->i_xattrs.index);
563
564 __free_xattr(xattr);
565 }
566
567 ci->i_xattrs.names_size = 0;
568 ci->i_xattrs.vals_size = 0;
569 ci->i_xattrs.index_version = 0;
570 ci->i_xattrs.count = 0;
571 ci->i_xattrs.index = RB_ROOT;
572}
573
574static int __build_xattrs(struct inode *inode)
575 __releases(ci->i_ceph_lock)
576 __acquires(ci->i_ceph_lock)
577{
578 u32 namelen;
579 u32 numattr = 0;
580 void *p, *end;
581 u32 len;
582 const char *name, *val;
583 struct ceph_inode_info *ci = ceph_inode(inode);
584 int xattr_version;
585 struct ceph_inode_xattr **xattrs = NULL;
586 int err = 0;
587 int i;
588
589 dout("__build_xattrs() len=%d\n",
590 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
591
592 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
593 return 0; /* already built */
594
595 __ceph_destroy_xattrs(ci);
596
597start:
598 /* updated internal xattr rb tree */
599 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
600 p = ci->i_xattrs.blob->vec.iov_base;
601 end = p + ci->i_xattrs.blob->vec.iov_len;
602 ceph_decode_32_safe(&p, end, numattr, bad);
603 xattr_version = ci->i_xattrs.version;
604 spin_unlock(&ci->i_ceph_lock);
605
606 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
607 GFP_NOFS);
608 err = -ENOMEM;
609 if (!xattrs)
610 goto bad_lock;
611
612 for (i = 0; i < numattr; i++) {
613 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
614 GFP_NOFS);
615 if (!xattrs[i])
616 goto bad_lock;
617 }
618
619 spin_lock(&ci->i_ceph_lock);
620 if (ci->i_xattrs.version != xattr_version) {
621 /* lost a race, retry */
622 for (i = 0; i < numattr; i++)
623 kfree(xattrs[i]);
624 kfree(xattrs);
625 xattrs = NULL;
626 goto start;
627 }
628 err = -EIO;
629 while (numattr--) {
630 ceph_decode_32_safe(&p, end, len, bad);
631 namelen = len;
632 name = p;
633 p += len;
634 ceph_decode_32_safe(&p, end, len, bad);
635 val = p;
636 p += len;
637
638 err = __set_xattr(ci, name, namelen, val, len,
639 0, 0, &xattrs[numattr]);
640
641 if (err < 0)
642 goto bad;
643 }
644 kfree(xattrs);
645 }
646 ci->i_xattrs.index_version = ci->i_xattrs.version;
647 ci->i_xattrs.dirty = false;
648
649 return err;
650bad_lock:
651 spin_lock(&ci->i_ceph_lock);
652bad:
653 if (xattrs) {
654 for (i = 0; i < numattr; i++)
655 kfree(xattrs[i]);
656 kfree(xattrs);
657 }
658 ci->i_xattrs.names_size = 0;
659 return err;
660}
661
662static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
663 int val_size)
664{
665 /*
666 * 4 bytes for the length, and additional 4 bytes per each xattr name,
667 * 4 bytes per each value
668 */
669 int size = 4 + ci->i_xattrs.count*(4 + 4) +
670 ci->i_xattrs.names_size +
671 ci->i_xattrs.vals_size;
672 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
673 ci->i_xattrs.count, ci->i_xattrs.names_size,
674 ci->i_xattrs.vals_size);
675
676 if (name_size)
677 size += 4 + 4 + name_size + val_size;
678
679 return size;
680}
681
682/*
683 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
684 * and swap into place. It returns the old i_xattrs.blob (or NULL) so
685 * that it can be freed by the caller as the i_ceph_lock is likely to be
686 * held.
687 */
688struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
689{
690 struct rb_node *p;
691 struct ceph_inode_xattr *xattr = NULL;
692 struct ceph_buffer *old_blob = NULL;
693 void *dest;
694
695 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
696 if (ci->i_xattrs.dirty) {
697 int need = __get_required_blob_size(ci, 0, 0);
698
699 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
700
701 p = rb_first(&ci->i_xattrs.index);
702 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
703
704 ceph_encode_32(&dest, ci->i_xattrs.count);
705 while (p) {
706 xattr = rb_entry(p, struct ceph_inode_xattr, node);
707
708 ceph_encode_32(&dest, xattr->name_len);
709 memcpy(dest, xattr->name, xattr->name_len);
710 dest += xattr->name_len;
711 ceph_encode_32(&dest, xattr->val_len);
712 memcpy(dest, xattr->val, xattr->val_len);
713 dest += xattr->val_len;
714
715 p = rb_next(p);
716 }
717
718 /* adjust buffer len; it may be larger than we need */
719 ci->i_xattrs.prealloc_blob->vec.iov_len =
720 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
721
722 if (ci->i_xattrs.blob)
723 old_blob = ci->i_xattrs.blob;
724 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
725 ci->i_xattrs.prealloc_blob = NULL;
726 ci->i_xattrs.dirty = false;
727 ci->i_xattrs.version++;
728 }
729
730 return old_blob;
731}
732
733static inline int __get_request_mask(struct inode *in) {
734 struct ceph_mds_request *req = current->journal_info;
735 int mask = 0;
736 if (req && req->r_target_inode == in) {
737 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
738 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
739 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
740 req->r_op == CEPH_MDS_OP_GETATTR) {
741 mask = le32_to_cpu(req->r_args.getattr.mask);
742 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
743 req->r_op == CEPH_MDS_OP_CREATE) {
744 mask = le32_to_cpu(req->r_args.open.mask);
745 }
746 }
747 return mask;
748}
749
750ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
751 size_t size)
752{
753 struct ceph_inode_info *ci = ceph_inode(inode);
754 struct ceph_inode_xattr *xattr;
755 struct ceph_vxattr *vxattr = NULL;
756 int req_mask;
757 int err;
758
759 /* let's see if a virtual xattr was requested */
760 vxattr = ceph_match_vxattr(inode, name);
761 if (vxattr) {
762 err = ceph_do_getattr(inode, 0, true);
763 if (err)
764 return err;
765 err = -ENODATA;
766 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
767 err = vxattr->getxattr_cb(ci, value, size);
768 if (size && size < err)
769 err = -ERANGE;
770 }
771 return err;
772 }
773
774 req_mask = __get_request_mask(inode);
775
776 spin_lock(&ci->i_ceph_lock);
777 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
778 ci->i_xattrs.version, ci->i_xattrs.index_version);
779
780 if (ci->i_xattrs.version == 0 ||
781 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
782 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
783 spin_unlock(&ci->i_ceph_lock);
784
785 /* security module gets xattr while filling trace */
786 if (current->journal_info) {
787 pr_warn_ratelimited("sync getxattr %p "
788 "during filling trace\n", inode);
789 return -EBUSY;
790 }
791
792 /* get xattrs from mds (if we don't already have them) */
793 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
794 if (err)
795 return err;
796 spin_lock(&ci->i_ceph_lock);
797 }
798
799 err = __build_xattrs(inode);
800 if (err < 0)
801 goto out;
802
803 err = -ENODATA; /* == ENOATTR */
804 xattr = __get_xattr(ci, name);
805 if (!xattr)
806 goto out;
807
808 err = -ERANGE;
809 if (size && size < xattr->val_len)
810 goto out;
811
812 err = xattr->val_len;
813 if (size == 0)
814 goto out;
815
816 memcpy(value, xattr->val, xattr->val_len);
817
818 if (current->journal_info &&
819 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
820 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
821out:
822 spin_unlock(&ci->i_ceph_lock);
823 return err;
824}
825
826ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
827{
828 struct inode *inode = d_inode(dentry);
829 struct ceph_inode_info *ci = ceph_inode(inode);
830 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
831 u32 vir_namelen = 0;
832 u32 namelen;
833 int err;
834 u32 len;
835 int i;
836
837 spin_lock(&ci->i_ceph_lock);
838 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
839 ci->i_xattrs.version, ci->i_xattrs.index_version);
840
841 if (ci->i_xattrs.version == 0 ||
842 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
843 spin_unlock(&ci->i_ceph_lock);
844 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
845 if (err)
846 return err;
847 spin_lock(&ci->i_ceph_lock);
848 }
849
850 err = __build_xattrs(inode);
851 if (err < 0)
852 goto out;
853 /*
854 * Start with virtual dir xattr names (if any) (including
855 * terminating '\0' characters for each).
856 */
857 vir_namelen = ceph_vxattrs_name_size(vxattrs);
858
859 /* adding 1 byte per each variable due to the null termination */
860 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
861 err = -ERANGE;
862 if (size && vir_namelen + namelen > size)
863 goto out;
864
865 err = namelen + vir_namelen;
866 if (size == 0)
867 goto out;
868
869 names = __copy_xattr_names(ci, names);
870
871 /* virtual xattr names, too */
872 err = namelen;
873 if (vxattrs) {
874 for (i = 0; vxattrs[i].name; i++) {
875 if (!vxattrs[i].hidden &&
876 !(vxattrs[i].exists_cb &&
877 !vxattrs[i].exists_cb(ci))) {
878 len = sprintf(names, "%s", vxattrs[i].name);
879 names += len + 1;
880 err += len + 1;
881 }
882 }
883 }
884
885out:
886 spin_unlock(&ci->i_ceph_lock);
887 return err;
888}
889
890static int ceph_sync_setxattr(struct inode *inode, const char *name,
891 const char *value, size_t size, int flags)
892{
893 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
894 struct ceph_inode_info *ci = ceph_inode(inode);
895 struct ceph_mds_request *req;
896 struct ceph_mds_client *mdsc = fsc->mdsc;
897 struct ceph_pagelist *pagelist = NULL;
898 int op = CEPH_MDS_OP_SETXATTR;
899 int err;
900
901 if (size > 0) {
902 /* copy value into pagelist */
903 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
904 if (!pagelist)
905 return -ENOMEM;
906
907 ceph_pagelist_init(pagelist);
908 err = ceph_pagelist_append(pagelist, value, size);
909 if (err)
910 goto out;
911 } else if (!value) {
912 if (flags & CEPH_XATTR_REPLACE)
913 op = CEPH_MDS_OP_RMXATTR;
914 else
915 flags |= CEPH_XATTR_REMOVE;
916 }
917
918 dout("setxattr value=%.*s\n", (int)size, value);
919
920 /* do request */
921 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
922 if (IS_ERR(req)) {
923 err = PTR_ERR(req);
924 goto out;
925 }
926
927 req->r_path2 = kstrdup(name, GFP_NOFS);
928 if (!req->r_path2) {
929 ceph_mdsc_put_request(req);
930 err = -ENOMEM;
931 goto out;
932 }
933
934 if (op == CEPH_MDS_OP_SETXATTR) {
935 req->r_args.setxattr.flags = cpu_to_le32(flags);
936 req->r_pagelist = pagelist;
937 pagelist = NULL;
938 }
939
940 req->r_inode = inode;
941 ihold(inode);
942 req->r_num_caps = 1;
943 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
944
945 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
946 err = ceph_mdsc_do_request(mdsc, NULL, req);
947 ceph_mdsc_put_request(req);
948 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
949
950out:
951 if (pagelist)
952 ceph_pagelist_release(pagelist);
953 return err;
954}
955
956int __ceph_setxattr(struct inode *inode, const char *name,
957 const void *value, size_t size, int flags)
958{
959 struct ceph_vxattr *vxattr;
960 struct ceph_inode_info *ci = ceph_inode(inode);
961 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
962 struct ceph_cap_flush *prealloc_cf = NULL;
963 struct ceph_buffer *old_blob = NULL;
964 int issued;
965 int err;
966 int dirty = 0;
967 int name_len = strlen(name);
968 int val_len = size;
969 char *newname = NULL;
970 char *newval = NULL;
971 struct ceph_inode_xattr *xattr = NULL;
972 int required_blob_size;
973 bool lock_snap_rwsem = false;
974
975 if (ceph_snap(inode) != CEPH_NOSNAP)
976 return -EROFS;
977
978 vxattr = ceph_match_vxattr(inode, name);
979 if (vxattr && vxattr->readonly)
980 return -EOPNOTSUPP;
981
982 /* pass any unhandled ceph.* xattrs through to the MDS */
983 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
984 goto do_sync_unlocked;
985
986 /* preallocate memory for xattr name, value, index node */
987 err = -ENOMEM;
988 newname = kmemdup(name, name_len + 1, GFP_NOFS);
989 if (!newname)
990 goto out;
991
992 if (val_len) {
993 newval = kmemdup(value, val_len, GFP_NOFS);
994 if (!newval)
995 goto out;
996 }
997
998 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
999 if (!xattr)
1000 goto out;
1001
1002 prealloc_cf = ceph_alloc_cap_flush();
1003 if (!prealloc_cf)
1004 goto out;
1005
1006 spin_lock(&ci->i_ceph_lock);
1007retry:
1008 issued = __ceph_caps_issued(ci, NULL);
1009 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1010 goto do_sync;
1011
1012 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1013 lock_snap_rwsem = true;
1014 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1015 spin_unlock(&ci->i_ceph_lock);
1016 down_read(&mdsc->snap_rwsem);
1017 spin_lock(&ci->i_ceph_lock);
1018 goto retry;
1019 }
1020 }
1021
1022 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1023 __build_xattrs(inode);
1024
1025 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1026
1027 if (!ci->i_xattrs.prealloc_blob ||
1028 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1029 struct ceph_buffer *blob;
1030
1031 spin_unlock(&ci->i_ceph_lock);
1032 ceph_buffer_put(old_blob); /* Shouldn't be required */
1033 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1034 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1035 if (!blob)
1036 goto do_sync_unlocked;
1037 spin_lock(&ci->i_ceph_lock);
1038 /* prealloc_blob can't be released while holding i_ceph_lock */
1039 if (ci->i_xattrs.prealloc_blob)
1040 old_blob = ci->i_xattrs.prealloc_blob;
1041 ci->i_xattrs.prealloc_blob = blob;
1042 goto retry;
1043 }
1044
1045 err = __set_xattr(ci, newname, name_len, newval, val_len,
1046 flags, value ? 1 : -1, &xattr);
1047
1048 if (!err) {
1049 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1050 &prealloc_cf);
1051 ci->i_xattrs.dirty = true;
1052 inode->i_ctime = current_time(inode);
1053 }
1054
1055 spin_unlock(&ci->i_ceph_lock);
1056 ceph_buffer_put(old_blob);
1057 if (lock_snap_rwsem)
1058 up_read(&mdsc->snap_rwsem);
1059 if (dirty)
1060 __mark_inode_dirty(inode, dirty);
1061 ceph_free_cap_flush(prealloc_cf);
1062 return err;
1063
1064do_sync:
1065 spin_unlock(&ci->i_ceph_lock);
1066do_sync_unlocked:
1067 if (lock_snap_rwsem)
1068 up_read(&mdsc->snap_rwsem);
1069
1070 /* security module set xattr while filling trace */
1071 if (current->journal_info) {
1072 pr_warn_ratelimited("sync setxattr %p "
1073 "during filling trace\n", inode);
1074 err = -EBUSY;
1075 } else {
1076 err = ceph_sync_setxattr(inode, name, value, size, flags);
1077 }
1078out:
1079 ceph_free_cap_flush(prealloc_cf);
1080 kfree(newname);
1081 kfree(newval);
1082 kfree(xattr);
1083 return err;
1084}
1085
1086static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1087 struct dentry *dentry, struct inode *inode,
1088 const char *name, void *value, size_t size)
1089{
1090 if (!ceph_is_valid_xattr(name))
1091 return -EOPNOTSUPP;
1092 return __ceph_getxattr(inode, name, value, size);
1093}
1094
1095static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1096 struct dentry *unused, struct inode *inode,
1097 const char *name, const void *value,
1098 size_t size, int flags)
1099{
1100 if (!ceph_is_valid_xattr(name))
1101 return -EOPNOTSUPP;
1102 return __ceph_setxattr(inode, name, value, size, flags);
1103}
1104
1105static const struct xattr_handler ceph_other_xattr_handler = {
1106 .prefix = "", /* match any name => handlers called with full name */
1107 .get = ceph_get_xattr_handler,
1108 .set = ceph_set_xattr_handler,
1109};
1110
1111#ifdef CONFIG_SECURITY
1112bool ceph_security_xattr_wanted(struct inode *in)
1113{
1114 return in->i_security != NULL;
1115}
1116
1117bool ceph_security_xattr_deadlock(struct inode *in)
1118{
1119 struct ceph_inode_info *ci;
1120 bool ret;
1121 if (!in->i_security)
1122 return false;
1123 ci = ceph_inode(in);
1124 spin_lock(&ci->i_ceph_lock);
1125 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1126 !(ci->i_xattrs.version > 0 &&
1127 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1128 spin_unlock(&ci->i_ceph_lock);
1129 return ret;
1130}
1131#endif