1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 #include <linux/ceph/pagelist.h>
6 #include "mds_client.h"
8 #include <linux/ceph/decode.h>
10 #include <linux/xattr.h>
11 #include <linux/security.h>
12 #include <linux/posix_acl_xattr.h>
13 #include <linux/slab.h>
15 #define XATTR_CEPH_PREFIX "ceph."
16 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
18 static int __remove_xattr(struct ceph_inode_info *ci,
19 struct ceph_inode_xattr *xattr);
21 static bool ceph_is_valid_xattr(const char *name)
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);
30 * These define virtual xattrs exposing the recursive directory
31 * statistics and layout metadata.
35 size_t name_size; /* strlen(name) + 1 (for '\0') */
36 ssize_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
38 bool (*exists_cb)(struct ceph_inode_info *ci);
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)
49 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
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);
57 static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
60 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
61 struct ceph_client *cl = fsc->client;
62 struct ceph_osd_client *osdc = &fsc->client->osdc;
63 struct ceph_string *pool_ns;
64 s64 pool = ci->i_layout.pool_id;
65 const char *pool_name;
66 const char *ns_field = " pool_namespace=";
68 size_t len, total_len = 0;
71 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
73 doutc(cl, "%p\n", &ci->netfs.inode);
74 down_read(&osdc->lock);
75 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
77 len = snprintf(buf, sizeof(buf),
78 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
79 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
80 ci->i_layout.object_size);
81 total_len = len + strlen(pool_name);
83 len = snprintf(buf, sizeof(buf),
84 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
85 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
86 ci->i_layout.object_size, pool);
91 total_len += strlen(ns_field) + pool_ns->len;
94 if (size >= total_len) {
95 memcpy(val, buf, len);
98 len = strlen(pool_name);
99 memcpy(val + ret, pool_name, len);
103 len = strlen(ns_field);
104 memcpy(val + ret, ns_field, len);
106 memcpy(val + ret, pool_ns->str, pool_ns->len);
110 up_read(&osdc->lock);
111 ceph_put_string(pool_ns);
116 * The convention with strings in xattrs is that they should not be NULL
117 * terminated, since we're returning the length with them. snprintf always
118 * NULL terminates however, so call it on a temporary buffer and then memcpy
119 * the result into place.
121 static __printf(3, 4)
122 int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
126 char buf[96]; /* NB: reevaluate size if new vxattrs are added */
129 ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
133 if (size && ret + 1 > sizeof(buf)) {
134 WARN_ONCE(true, "Returned length too big (%d)", ret);
139 memcpy(val, buf, ret);
143 static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
144 char *val, size_t size)
146 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
149 static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
150 char *val, size_t size)
152 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
155 static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
156 char *val, size_t size)
158 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
161 static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
162 char *val, size_t size)
165 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
166 struct ceph_osd_client *osdc = &fsc->client->osdc;
167 s64 pool = ci->i_layout.pool_id;
168 const char *pool_name;
170 down_read(&osdc->lock);
171 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
173 ret = strlen(pool_name);
175 memcpy(val, pool_name, ret);
177 ret = ceph_fmt_xattr(val, size, "%lld", pool);
179 up_read(&osdc->lock);
183 static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
184 char *val, size_t size)
187 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
192 memcpy(val, ns->str, ret);
200 static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
203 return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
206 static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
209 return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
212 static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
215 return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
218 static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
221 return ceph_fmt_xattr(val, size, "%lld",
222 ci->i_rfiles + ci->i_rsubdirs);
225 static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
228 return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
231 static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
234 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
237 static ssize_t ceph_vxattrcb_dir_rsnaps(struct ceph_inode_info *ci, char *val,
240 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsnaps);
243 static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
246 return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
249 static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
252 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
253 ci->i_rctime.tv_nsec);
257 static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
259 return ci->i_dir_pin != -ENODATA;
262 static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
265 return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
269 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
272 spin_lock(&ci->i_ceph_lock);
273 if ((ci->i_max_files || ci->i_max_bytes) &&
274 ci->i_vino.snap == CEPH_NOSNAP &&
276 ci->i_snap_realm->ino == ci->i_vino.ino)
278 spin_unlock(&ci->i_ceph_lock);
282 static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
285 return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
286 ci->i_max_bytes, ci->i_max_files);
289 static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
290 char *val, size_t size)
292 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
295 static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
296 char *val, size_t size)
298 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
302 static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
304 return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
307 static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
310 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
311 ci->i_snap_btime.tv_nsec);
314 static ssize_t ceph_vxattrcb_cluster_fsid(struct ceph_inode_info *ci,
315 char *val, size_t size)
317 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
319 return ceph_fmt_xattr(val, size, "%pU", &fsc->client->fsid);
322 static ssize_t ceph_vxattrcb_client_id(struct ceph_inode_info *ci,
323 char *val, size_t size)
325 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
327 return ceph_fmt_xattr(val, size, "client%lld",
328 ceph_client_gid(fsc->client));
331 static ssize_t ceph_vxattrcb_caps(struct ceph_inode_info *ci, char *val,
336 spin_lock(&ci->i_ceph_lock);
337 issued = __ceph_caps_issued(ci, NULL);
338 spin_unlock(&ci->i_ceph_lock);
340 return ceph_fmt_xattr(val, size, "%s/0x%x",
341 ceph_cap_string(issued), issued);
344 static ssize_t ceph_vxattrcb_auth_mds(struct ceph_inode_info *ci,
345 char *val, size_t size)
349 spin_lock(&ci->i_ceph_lock);
350 ret = ceph_fmt_xattr(val, size, "%d",
351 ci->i_auth_cap ? ci->i_auth_cap->session->s_mds : -1);
352 spin_unlock(&ci->i_ceph_lock);
356 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
357 static bool ceph_vxattrcb_fscrypt_auth_exists(struct ceph_inode_info *ci)
359 return ci->fscrypt_auth_len;
362 static ssize_t ceph_vxattrcb_fscrypt_auth(struct ceph_inode_info *ci,
363 char *val, size_t size)
366 if (size < ci->fscrypt_auth_len)
368 memcpy(val, ci->fscrypt_auth, ci->fscrypt_auth_len);
370 return ci->fscrypt_auth_len;
372 #endif /* CONFIG_FS_ENCRYPTION */
374 #define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
375 #define CEPH_XATTR_NAME2(_type, _name, _name2) \
376 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
378 #define XATTR_NAME_CEPH(_type, _name, _flags) \
380 .name = CEPH_XATTR_NAME(_type, _name), \
381 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
382 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
384 .flags = (VXATTR_FLAG_READONLY | _flags), \
386 #define XATTR_RSTAT_FIELD(_type, _name) \
387 XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
388 #define XATTR_RSTAT_FIELD_UPDATABLE(_type, _name) \
390 .name = CEPH_XATTR_NAME(_type, _name), \
391 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
392 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
394 .flags = VXATTR_FLAG_RSTAT, \
396 #define XATTR_LAYOUT_FIELD(_type, _name, _field) \
398 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
399 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
400 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
401 .exists_cb = ceph_vxattrcb_layout_exists, \
402 .flags = VXATTR_FLAG_HIDDEN, \
404 #define XATTR_QUOTA_FIELD(_type, _name) \
406 .name = CEPH_XATTR_NAME(_type, _name), \
407 .name_size = sizeof(CEPH_XATTR_NAME(_type, _name)), \
408 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
409 .exists_cb = ceph_vxattrcb_quota_exists, \
410 .flags = VXATTR_FLAG_HIDDEN, \
413 static struct ceph_vxattr ceph_dir_vxattrs[] = {
415 .name = "ceph.dir.layout",
416 .name_size = sizeof("ceph.dir.layout"),
417 .getxattr_cb = ceph_vxattrcb_layout,
418 .exists_cb = ceph_vxattrcb_layout_exists,
419 .flags = VXATTR_FLAG_HIDDEN,
421 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
422 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
423 XATTR_LAYOUT_FIELD(dir, layout, object_size),
424 XATTR_LAYOUT_FIELD(dir, layout, pool),
425 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
426 XATTR_NAME_CEPH(dir, entries, VXATTR_FLAG_DIRSTAT),
427 XATTR_NAME_CEPH(dir, files, VXATTR_FLAG_DIRSTAT),
428 XATTR_NAME_CEPH(dir, subdirs, VXATTR_FLAG_DIRSTAT),
429 XATTR_RSTAT_FIELD(dir, rentries),
430 XATTR_RSTAT_FIELD(dir, rfiles),
431 XATTR_RSTAT_FIELD(dir, rsubdirs),
432 XATTR_RSTAT_FIELD(dir, rsnaps),
433 XATTR_RSTAT_FIELD(dir, rbytes),
434 XATTR_RSTAT_FIELD_UPDATABLE(dir, rctime),
436 .name = "ceph.dir.pin",
437 .name_size = sizeof("ceph.dir.pin"),
438 .getxattr_cb = ceph_vxattrcb_dir_pin,
439 .exists_cb = ceph_vxattrcb_dir_pin_exists,
440 .flags = VXATTR_FLAG_HIDDEN,
443 .name = "ceph.quota",
444 .name_size = sizeof("ceph.quota"),
445 .getxattr_cb = ceph_vxattrcb_quota,
446 .exists_cb = ceph_vxattrcb_quota_exists,
447 .flags = VXATTR_FLAG_HIDDEN,
449 XATTR_QUOTA_FIELD(quota, max_bytes),
450 XATTR_QUOTA_FIELD(quota, max_files),
452 .name = "ceph.snap.btime",
453 .name_size = sizeof("ceph.snap.btime"),
454 .getxattr_cb = ceph_vxattrcb_snap_btime,
455 .exists_cb = ceph_vxattrcb_snap_btime_exists,
456 .flags = VXATTR_FLAG_READONLY,
460 .name_size = sizeof("ceph.caps"),
461 .getxattr_cb = ceph_vxattrcb_caps,
463 .flags = VXATTR_FLAG_HIDDEN,
465 { .name = NULL, 0 } /* Required table terminator */
470 static struct ceph_vxattr ceph_file_vxattrs[] = {
472 .name = "ceph.file.layout",
473 .name_size = sizeof("ceph.file.layout"),
474 .getxattr_cb = ceph_vxattrcb_layout,
475 .exists_cb = ceph_vxattrcb_layout_exists,
476 .flags = VXATTR_FLAG_HIDDEN,
478 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
479 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
480 XATTR_LAYOUT_FIELD(file, layout, object_size),
481 XATTR_LAYOUT_FIELD(file, layout, pool),
482 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
484 .name = "ceph.snap.btime",
485 .name_size = sizeof("ceph.snap.btime"),
486 .getxattr_cb = ceph_vxattrcb_snap_btime,
487 .exists_cb = ceph_vxattrcb_snap_btime_exists,
488 .flags = VXATTR_FLAG_READONLY,
492 .name_size = sizeof("ceph.caps"),
493 .getxattr_cb = ceph_vxattrcb_caps,
495 .flags = VXATTR_FLAG_HIDDEN,
497 { .name = NULL, 0 } /* Required table terminator */
500 static struct ceph_vxattr ceph_common_vxattrs[] = {
502 .name = "ceph.cluster_fsid",
503 .name_size = sizeof("ceph.cluster_fsid"),
504 .getxattr_cb = ceph_vxattrcb_cluster_fsid,
506 .flags = VXATTR_FLAG_READONLY,
509 .name = "ceph.client_id",
510 .name_size = sizeof("ceph.client_id"),
511 .getxattr_cb = ceph_vxattrcb_client_id,
513 .flags = VXATTR_FLAG_READONLY,
516 .name = "ceph.auth_mds",
517 .name_size = sizeof("ceph.auth_mds"),
518 .getxattr_cb = ceph_vxattrcb_auth_mds,
520 .flags = VXATTR_FLAG_READONLY,
522 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
524 .name = "ceph.fscrypt.auth",
525 .name_size = sizeof("ceph.fscrypt.auth"),
526 .getxattr_cb = ceph_vxattrcb_fscrypt_auth,
527 .exists_cb = ceph_vxattrcb_fscrypt_auth_exists,
528 .flags = VXATTR_FLAG_READONLY,
530 #endif /* CONFIG_FS_ENCRYPTION */
531 { .name = NULL, 0 } /* Required table terminator */
534 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
536 if (S_ISDIR(inode->i_mode))
537 return ceph_dir_vxattrs;
538 else if (S_ISREG(inode->i_mode))
539 return ceph_file_vxattrs;
543 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
546 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
549 while (vxattr->name) {
550 if (!strcmp(vxattr->name, name))
556 vxattr = ceph_common_vxattrs;
557 while (vxattr->name) {
558 if (!strcmp(vxattr->name, name))
566 #define MAX_XATTR_VAL_PRINT_LEN 256
568 static int __set_xattr(struct ceph_inode_info *ci,
569 const char *name, int name_len,
570 const char *val, int val_len,
571 int flags, int update_xattr,
572 struct ceph_inode_xattr **newxattr)
574 struct inode *inode = &ci->netfs.inode;
575 struct ceph_client *cl = ceph_inode_to_client(inode);
577 struct rb_node *parent = NULL;
578 struct ceph_inode_xattr *xattr = NULL;
582 p = &ci->i_xattrs.index.rb_node;
585 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
586 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
592 if (name_len == xattr->name_len)
594 else if (name_len < xattr->name_len)
605 if (xattr && (flags & XATTR_CREATE))
607 else if (!xattr && (flags & XATTR_REPLACE))
615 if (update_xattr < 0) {
617 __remove_xattr(ci, xattr);
628 xattr->name_len = name_len;
629 xattr->should_free_name = update_xattr;
631 ci->i_xattrs.count++;
632 doutc(cl, "count=%d\n", ci->i_xattrs.count);
636 if (xattr->should_free_val)
643 ci->i_xattrs.names_size -= xattr->name_len;
644 ci->i_xattrs.vals_size -= xattr->val_len;
646 ci->i_xattrs.names_size += name_len;
647 ci->i_xattrs.vals_size += val_len;
653 xattr->val_len = val_len;
654 xattr->dirty = update_xattr;
655 xattr->should_free_val = (val && update_xattr);
658 rb_link_node(&xattr->node, parent, p);
659 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
660 doutc(cl, "p=%p\n", p);
663 doutc(cl, "added %p %llx.%llx xattr %p %.*s=%.*s%s\n", inode,
664 ceph_vinop(inode), xattr, name_len, name, min(val_len,
665 MAX_XATTR_VAL_PRINT_LEN), val,
666 val_len > MAX_XATTR_VAL_PRINT_LEN ? "..." : "");
671 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
674 struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
676 struct rb_node *parent = NULL;
677 struct ceph_inode_xattr *xattr = NULL;
678 int name_len = strlen(name);
681 p = &ci->i_xattrs.index.rb_node;
684 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
685 c = strncmp(name, xattr->name, xattr->name_len);
686 if (c == 0 && name_len > xattr->name_len)
693 int len = min(xattr->val_len, MAX_XATTR_VAL_PRINT_LEN);
695 doutc(cl, "%s found %.*s%s\n", name, len, xattr->val,
696 xattr->val_len > len ? "..." : "");
701 doutc(cl, "%s not found\n", name);
706 static void __free_xattr(struct ceph_inode_xattr *xattr)
710 if (xattr->should_free_name)
712 if (xattr->should_free_val)
718 static int __remove_xattr(struct ceph_inode_info *ci,
719 struct ceph_inode_xattr *xattr)
724 rb_erase(&xattr->node, &ci->i_xattrs.index);
726 if (xattr->should_free_name)
728 if (xattr->should_free_val)
731 ci->i_xattrs.names_size -= xattr->name_len;
732 ci->i_xattrs.vals_size -= xattr->val_len;
733 ci->i_xattrs.count--;
739 static char *__copy_xattr_names(struct ceph_inode_info *ci,
742 struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
744 struct ceph_inode_xattr *xattr = NULL;
746 p = rb_first(&ci->i_xattrs.index);
747 doutc(cl, "count=%d\n", ci->i_xattrs.count);
750 xattr = rb_entry(p, struct ceph_inode_xattr, node);
751 memcpy(dest, xattr->name, xattr->name_len);
752 dest[xattr->name_len] = '\0';
754 doutc(cl, "dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
755 xattr->name_len, ci->i_xattrs.names_size);
757 dest += xattr->name_len + 1;
764 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
766 struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
767 struct rb_node *p, *tmp;
768 struct ceph_inode_xattr *xattr = NULL;
770 p = rb_first(&ci->i_xattrs.index);
772 doutc(cl, "p=%p\n", p);
775 xattr = rb_entry(p, struct ceph_inode_xattr, node);
778 doutc(cl, "next p=%p (%.*s)\n", p, xattr->name_len, xattr->name);
779 rb_erase(tmp, &ci->i_xattrs.index);
784 ci->i_xattrs.names_size = 0;
785 ci->i_xattrs.vals_size = 0;
786 ci->i_xattrs.index_version = 0;
787 ci->i_xattrs.count = 0;
788 ci->i_xattrs.index = RB_ROOT;
791 static int __build_xattrs(struct inode *inode)
792 __releases(ci->i_ceph_lock)
793 __acquires(ci->i_ceph_lock)
795 struct ceph_client *cl = ceph_inode_to_client(inode);
800 const char *name, *val;
801 struct ceph_inode_info *ci = ceph_inode(inode);
803 struct ceph_inode_xattr **xattrs = NULL;
807 doutc(cl, "len=%d\n",
808 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
810 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
811 return 0; /* already built */
813 __ceph_destroy_xattrs(ci);
816 /* updated internal xattr rb tree */
817 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
818 p = ci->i_xattrs.blob->vec.iov_base;
819 end = p + ci->i_xattrs.blob->vec.iov_len;
820 ceph_decode_32_safe(&p, end, numattr, bad);
821 xattr_version = ci->i_xattrs.version;
822 spin_unlock(&ci->i_ceph_lock);
824 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
830 for (i = 0; i < numattr; i++) {
831 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
837 spin_lock(&ci->i_ceph_lock);
838 if (ci->i_xattrs.version != xattr_version) {
839 /* lost a race, retry */
840 for (i = 0; i < numattr; i++)
848 ceph_decode_32_safe(&p, end, len, bad);
852 ceph_decode_32_safe(&p, end, len, bad);
856 err = __set_xattr(ci, name, namelen, val, len,
857 0, 0, &xattrs[numattr]);
864 ci->i_xattrs.index_version = ci->i_xattrs.version;
865 ci->i_xattrs.dirty = false;
869 spin_lock(&ci->i_ceph_lock);
872 for (i = 0; i < numattr; i++)
876 ci->i_xattrs.names_size = 0;
880 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
883 struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
886 * 4 bytes for the length, and additional 4 bytes per each xattr name,
887 * 4 bytes per each value
889 int size = 4 + ci->i_xattrs.count*(4 + 4) +
890 ci->i_xattrs.names_size +
891 ci->i_xattrs.vals_size;
892 doutc(cl, "c=%d names.size=%d vals.size=%d\n", ci->i_xattrs.count,
893 ci->i_xattrs.names_size, ci->i_xattrs.vals_size);
896 size += 4 + 4 + name_size + val_size;
902 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
903 * and swap into place. It returns the old i_xattrs.blob (or NULL) so
904 * that it can be freed by the caller as the i_ceph_lock is likely to be
907 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
909 struct inode *inode = &ci->netfs.inode;
910 struct ceph_client *cl = ceph_inode_to_client(inode);
912 struct ceph_inode_xattr *xattr = NULL;
913 struct ceph_buffer *old_blob = NULL;
916 doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
917 if (ci->i_xattrs.dirty) {
918 int need = __get_required_blob_size(ci, 0, 0);
920 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
922 p = rb_first(&ci->i_xattrs.index);
923 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
925 ceph_encode_32(&dest, ci->i_xattrs.count);
927 xattr = rb_entry(p, struct ceph_inode_xattr, node);
929 ceph_encode_32(&dest, xattr->name_len);
930 memcpy(dest, xattr->name, xattr->name_len);
931 dest += xattr->name_len;
932 ceph_encode_32(&dest, xattr->val_len);
933 memcpy(dest, xattr->val, xattr->val_len);
934 dest += xattr->val_len;
939 /* adjust buffer len; it may be larger than we need */
940 ci->i_xattrs.prealloc_blob->vec.iov_len =
941 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
943 if (ci->i_xattrs.blob)
944 old_blob = ci->i_xattrs.blob;
945 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
946 ci->i_xattrs.prealloc_blob = NULL;
947 ci->i_xattrs.dirty = false;
948 ci->i_xattrs.version++;
954 static inline int __get_request_mask(struct inode *in) {
955 struct ceph_mds_request *req = current->journal_info;
957 if (req && req->r_target_inode == in) {
958 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
959 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
960 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
961 req->r_op == CEPH_MDS_OP_GETATTR) {
962 mask = le32_to_cpu(req->r_args.getattr.mask);
963 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
964 req->r_op == CEPH_MDS_OP_CREATE) {
965 mask = le32_to_cpu(req->r_args.open.mask);
971 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
974 struct ceph_client *cl = ceph_inode_to_client(inode);
975 struct ceph_inode_info *ci = ceph_inode(inode);
976 struct ceph_inode_xattr *xattr;
977 struct ceph_vxattr *vxattr;
981 if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
982 goto handle_non_vxattrs;
984 /* let's see if a virtual xattr was requested */
985 vxattr = ceph_match_vxattr(inode, name);
988 if (vxattr->flags & VXATTR_FLAG_RSTAT)
989 mask |= CEPH_STAT_RSTAT;
990 if (vxattr->flags & VXATTR_FLAG_DIRSTAT)
991 mask |= CEPH_CAP_FILE_SHARED;
992 err = ceph_do_getattr(inode, mask, true);
996 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
997 err = vxattr->getxattr_cb(ci, value, size);
998 if (size && size < err)
1003 err = ceph_do_getvxattr(inode, name, value, size);
1004 /* this would happen with a new client and old server combo */
1005 if (err == -EOPNOTSUPP)
1010 req_mask = __get_request_mask(inode);
1012 spin_lock(&ci->i_ceph_lock);
1013 doutc(cl, "%p %llx.%llx name '%s' ver=%lld index_ver=%lld\n", inode,
1014 ceph_vinop(inode), name, ci->i_xattrs.version,
1015 ci->i_xattrs.index_version);
1017 if (ci->i_xattrs.version == 0 ||
1018 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
1019 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1))) {
1020 spin_unlock(&ci->i_ceph_lock);
1022 /* security module gets xattr while filling trace */
1023 if (current->journal_info) {
1024 pr_warn_ratelimited_client(cl,
1025 "sync %p %llx.%llx during filling trace\n",
1026 inode, ceph_vinop(inode));
1030 /* get xattrs from mds (if we don't already have them) */
1031 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1034 spin_lock(&ci->i_ceph_lock);
1037 err = __build_xattrs(inode);
1041 err = -ENODATA; /* == ENOATTR */
1042 xattr = __get_xattr(ci, name);
1047 if (size && size < xattr->val_len)
1050 err = xattr->val_len;
1054 memcpy(value, xattr->val, xattr->val_len);
1056 if (current->journal_info &&
1057 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
1058 security_ismaclabel(name + XATTR_SECURITY_PREFIX_LEN))
1059 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
1061 spin_unlock(&ci->i_ceph_lock);
1065 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1067 struct inode *inode = d_inode(dentry);
1068 struct ceph_client *cl = ceph_inode_to_client(inode);
1069 struct ceph_inode_info *ci = ceph_inode(inode);
1070 bool len_only = (size == 0);
1074 spin_lock(&ci->i_ceph_lock);
1075 doutc(cl, "%p %llx.%llx ver=%lld index_ver=%lld\n", inode,
1076 ceph_vinop(inode), ci->i_xattrs.version,
1077 ci->i_xattrs.index_version);
1079 if (ci->i_xattrs.version == 0 ||
1080 !__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 1)) {
1081 spin_unlock(&ci->i_ceph_lock);
1082 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
1085 spin_lock(&ci->i_ceph_lock);
1088 err = __build_xattrs(inode);
1092 /* add 1 byte for each xattr due to the null termination */
1093 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1095 if (namelen > size) {
1099 names = __copy_xattr_names(ci, names);
1104 spin_unlock(&ci->i_ceph_lock);
1108 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1109 const char *value, size_t size, int flags)
1111 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
1112 struct ceph_client *cl = ceph_inode_to_client(inode);
1113 struct ceph_inode_info *ci = ceph_inode(inode);
1114 struct ceph_mds_request *req;
1115 struct ceph_mds_client *mdsc = fsc->mdsc;
1116 struct ceph_osd_client *osdc = &fsc->client->osdc;
1117 struct ceph_pagelist *pagelist = NULL;
1118 int op = CEPH_MDS_OP_SETXATTR;
1122 /* copy value into pagelist */
1123 pagelist = ceph_pagelist_alloc(GFP_NOFS);
1127 err = ceph_pagelist_append(pagelist, value, size);
1130 } else if (!value) {
1131 if (flags & CEPH_XATTR_REPLACE)
1132 op = CEPH_MDS_OP_RMXATTR;
1134 flags |= CEPH_XATTR_REMOVE;
1137 doutc(cl, "name %s value size %zu\n", name, size);
1140 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1146 req->r_path2 = kstrdup(name, GFP_NOFS);
1147 if (!req->r_path2) {
1148 ceph_mdsc_put_request(req);
1153 if (op == CEPH_MDS_OP_SETXATTR) {
1154 req->r_args.setxattr.flags = cpu_to_le32(flags);
1155 req->r_args.setxattr.osdmap_epoch =
1156 cpu_to_le32(osdc->osdmap->epoch);
1157 req->r_pagelist = pagelist;
1161 req->r_inode = inode;
1163 req->r_num_caps = 1;
1164 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1166 doutc(cl, "xattr.ver (before): %lld\n", ci->i_xattrs.version);
1167 err = ceph_mdsc_do_request(mdsc, NULL, req);
1168 ceph_mdsc_put_request(req);
1169 doutc(cl, "xattr.ver (after): %lld\n", ci->i_xattrs.version);
1173 ceph_pagelist_release(pagelist);
1177 int __ceph_setxattr(struct inode *inode, const char *name,
1178 const void *value, size_t size, int flags)
1180 struct ceph_client *cl = ceph_inode_to_client(inode);
1181 struct ceph_vxattr *vxattr;
1182 struct ceph_inode_info *ci = ceph_inode(inode);
1183 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
1184 struct ceph_cap_flush *prealloc_cf = NULL;
1185 struct ceph_buffer *old_blob = NULL;
1189 int name_len = strlen(name);
1191 char *newname = NULL;
1192 char *newval = NULL;
1193 struct ceph_inode_xattr *xattr = NULL;
1194 int required_blob_size;
1195 bool check_realm = false;
1196 bool lock_snap_rwsem = false;
1198 if (ceph_snap(inode) != CEPH_NOSNAP)
1201 vxattr = ceph_match_vxattr(inode, name);
1203 if (vxattr->flags & VXATTR_FLAG_READONLY)
1205 if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1209 /* pass any unhandled ceph.* xattrs through to the MDS */
1210 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1211 goto do_sync_unlocked;
1213 /* preallocate memory for xattr name, value, index node */
1215 newname = kmemdup(name, name_len + 1, GFP_NOFS);
1220 newval = kmemdup(value, val_len, GFP_NOFS);
1225 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1229 prealloc_cf = ceph_alloc_cap_flush();
1233 spin_lock(&ci->i_ceph_lock);
1235 issued = __ceph_caps_issued(ci, NULL);
1236 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1237 if ((ci->i_xattrs.version == 0) || !(issued & CEPH_CAP_XATTR_EXCL) ||
1238 (required_blob_size > mdsc->mdsmap->m_max_xattr_size)) {
1239 doutc(cl, "sync version: %llu size: %d max: %llu\n",
1240 ci->i_xattrs.version, required_blob_size,
1241 mdsc->mdsmap->m_max_xattr_size);
1245 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1246 lock_snap_rwsem = true;
1247 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1248 spin_unlock(&ci->i_ceph_lock);
1249 down_read(&mdsc->snap_rwsem);
1250 spin_lock(&ci->i_ceph_lock);
1255 doutc(cl, "%p %llx.%llx name '%s' issued %s\n", inode,
1256 ceph_vinop(inode), name, ceph_cap_string(issued));
1257 __build_xattrs(inode);
1259 if (!ci->i_xattrs.prealloc_blob ||
1260 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1261 struct ceph_buffer *blob;
1263 spin_unlock(&ci->i_ceph_lock);
1264 ceph_buffer_put(old_blob); /* Shouldn't be required */
1265 doutc(cl, " pre-allocating new blob size=%d\n",
1266 required_blob_size);
1267 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1269 goto do_sync_unlocked;
1270 spin_lock(&ci->i_ceph_lock);
1271 /* prealloc_blob can't be released while holding i_ceph_lock */
1272 if (ci->i_xattrs.prealloc_blob)
1273 old_blob = ci->i_xattrs.prealloc_blob;
1274 ci->i_xattrs.prealloc_blob = blob;
1278 err = __set_xattr(ci, newname, name_len, newval, val_len,
1279 flags, value ? 1 : -1, &xattr);
1282 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1284 ci->i_xattrs.dirty = true;
1285 inode_set_ctime_current(inode);
1288 spin_unlock(&ci->i_ceph_lock);
1289 ceph_buffer_put(old_blob);
1290 if (lock_snap_rwsem)
1291 up_read(&mdsc->snap_rwsem);
1293 __mark_inode_dirty(inode, dirty);
1294 ceph_free_cap_flush(prealloc_cf);
1298 spin_unlock(&ci->i_ceph_lock);
1300 if (lock_snap_rwsem)
1301 up_read(&mdsc->snap_rwsem);
1303 /* security module set xattr while filling trace */
1304 if (current->journal_info) {
1305 pr_warn_ratelimited_client(cl,
1306 "sync %p %llx.%llx during filling trace\n",
1307 inode, ceph_vinop(inode));
1310 err = ceph_sync_setxattr(inode, name, value, size, flags);
1311 if (err >= 0 && check_realm) {
1312 /* check if snaprealm was created for quota inode */
1313 spin_lock(&ci->i_ceph_lock);
1314 if ((ci->i_max_files || ci->i_max_bytes) &&
1315 !(ci->i_snap_realm &&
1316 ci->i_snap_realm->ino == ci->i_vino.ino))
1318 spin_unlock(&ci->i_ceph_lock);
1322 ceph_free_cap_flush(prealloc_cf);
1329 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1330 struct dentry *dentry, struct inode *inode,
1331 const char *name, void *value, size_t size)
1333 if (!ceph_is_valid_xattr(name))
1335 return __ceph_getxattr(inode, name, value, size);
1338 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1339 struct mnt_idmap *idmap,
1340 struct dentry *unused, struct inode *inode,
1341 const char *name, const void *value,
1342 size_t size, int flags)
1344 if (!ceph_is_valid_xattr(name))
1346 return __ceph_setxattr(inode, name, value, size, flags);
1349 static const struct xattr_handler ceph_other_xattr_handler = {
1350 .prefix = "", /* match any name => handlers called with full name */
1351 .get = ceph_get_xattr_handler,
1352 .set = ceph_set_xattr_handler,
1355 #ifdef CONFIG_SECURITY
1356 bool ceph_security_xattr_wanted(struct inode *in)
1358 return in->i_security != NULL;
1361 bool ceph_security_xattr_deadlock(struct inode *in)
1363 struct ceph_inode_info *ci;
1365 if (!in->i_security)
1367 ci = ceph_inode(in);
1368 spin_lock(&ci->i_ceph_lock);
1369 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1370 !(ci->i_xattrs.version > 0 &&
1371 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1372 spin_unlock(&ci->i_ceph_lock);
1376 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1377 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1378 struct ceph_acl_sec_ctx *as_ctx)
1380 struct ceph_pagelist *pagelist = as_ctx->pagelist;
1385 err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1386 &name, &as_ctx->sec_ctx,
1387 &as_ctx->sec_ctxlen);
1389 WARN_ON_ONCE(err != -EOPNOTSUPP);
1390 err = 0; /* do nothing */
1396 pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1399 err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1402 ceph_pagelist_encode_32(pagelist, 1);
1406 * FIXME: Make security_dentry_init_security() generic. Currently
1407 * It only supports single security module and only selinux has
1408 * dentry_init_security hook.
1410 name_len = strlen(name);
1411 err = ceph_pagelist_reserve(pagelist,
1412 4 * 2 + name_len + as_ctx->sec_ctxlen);
1416 if (as_ctx->pagelist) {
1417 /* update count of KV pairs */
1418 BUG_ON(pagelist->length <= sizeof(__le32));
1419 if (list_is_singular(&pagelist->head)) {
1420 le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1422 struct page *page = list_first_entry(&pagelist->head,
1424 void *addr = kmap_atomic(page);
1425 le32_add_cpu((__le32*)addr, 1);
1426 kunmap_atomic(addr);
1429 as_ctx->pagelist = pagelist;
1432 ceph_pagelist_encode_32(pagelist, name_len);
1433 ceph_pagelist_append(pagelist, name, name_len);
1435 ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1436 ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1440 if (pagelist && !as_ctx->pagelist)
1441 ceph_pagelist_release(pagelist);
1444 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1445 #endif /* CONFIG_SECURITY */
1447 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1449 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1450 posix_acl_release(as_ctx->acl);
1451 posix_acl_release(as_ctx->default_acl);
1453 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1454 security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1456 #ifdef CONFIG_FS_ENCRYPTION
1457 kfree(as_ctx->fscrypt_auth);
1459 if (as_ctx->pagelist)
1460 ceph_pagelist_release(as_ctx->pagelist);
1464 * List of handlers for synthetic system.* attributes. Other
1465 * attributes are handled directly.
1467 const struct xattr_handler * const ceph_xattr_handlers[] = {
1468 &ceph_other_xattr_handler,