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_client(ci->vfs_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=";
67 size_t len, total_len = 0;
70 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
72 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
73 down_read(&osdc->lock);
74 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
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);
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);
90 total_len += strlen(ns_field) + pool_ns->len;
93 if (size >= total_len) {
94 memcpy(val, buf, len);
97 len = strlen(pool_name);
98 memcpy(val + ret, pool_name, len);
102 len = strlen(ns_field);
103 memcpy(val + ret, ns_field, len);
105 memcpy(val + ret, pool_ns->str, pool_ns->len);
109 up_read(&osdc->lock);
110 ceph_put_string(pool_ns);
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.
120 static __printf(3, 4)
121 int ceph_fmt_xattr(char *val, size_t size, const char *fmt, ...)
125 char buf[96]; /* NB: reevaluate size if new vxattrs are added */
128 ret = vsnprintf(buf, size ? sizeof(buf) : 0, fmt, args);
132 if (size && ret + 1 > sizeof(buf)) {
133 WARN_ONCE(true, "Returned length too big (%d)", ret);
138 memcpy(val, buf, ret);
142 static ssize_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
143 char *val, size_t size)
145 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_unit);
148 static ssize_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
149 char *val, size_t size)
151 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.stripe_count);
154 static ssize_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
155 char *val, size_t size)
157 return ceph_fmt_xattr(val, size, "%u", ci->i_layout.object_size);
160 static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
161 char *val, size_t size)
164 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_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;
169 down_read(&osdc->lock);
170 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
172 ret = strlen(pool_name);
174 memcpy(val, pool_name, ret);
176 ret = ceph_fmt_xattr(val, size, "%lld", pool);
178 up_read(&osdc->lock);
182 static ssize_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
183 char *val, size_t size)
186 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
191 memcpy(val, ns->str, ret);
199 static ssize_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
202 return ceph_fmt_xattr(val, size, "%lld", ci->i_files + ci->i_subdirs);
205 static ssize_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
208 return ceph_fmt_xattr(val, size, "%lld", ci->i_files);
211 static ssize_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
214 return ceph_fmt_xattr(val, size, "%lld", ci->i_subdirs);
217 static ssize_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
220 return ceph_fmt_xattr(val, size, "%lld",
221 ci->i_rfiles + ci->i_rsubdirs);
224 static ssize_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
227 return ceph_fmt_xattr(val, size, "%lld", ci->i_rfiles);
230 static ssize_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
233 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsubdirs);
236 static ssize_t ceph_vxattrcb_dir_rsnaps(struct ceph_inode_info *ci, char *val,
239 return ceph_fmt_xattr(val, size, "%lld", ci->i_rsnaps);
242 static ssize_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
245 return ceph_fmt_xattr(val, size, "%lld", ci->i_rbytes);
248 static ssize_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
251 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
252 ci->i_rctime.tv_nsec);
256 static bool ceph_vxattrcb_dir_pin_exists(struct ceph_inode_info *ci)
258 return ci->i_dir_pin != -ENODATA;
261 static ssize_t ceph_vxattrcb_dir_pin(struct ceph_inode_info *ci, char *val,
264 return ceph_fmt_xattr(val, size, "%d", (int)ci->i_dir_pin);
268 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
271 spin_lock(&ci->i_ceph_lock);
272 if ((ci->i_max_files || ci->i_max_bytes) &&
273 ci->i_vino.snap == CEPH_NOSNAP &&
275 ci->i_snap_realm->ino == ci->i_vino.ino)
277 spin_unlock(&ci->i_ceph_lock);
281 static ssize_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
284 return ceph_fmt_xattr(val, size, "max_bytes=%llu max_files=%llu",
285 ci->i_max_bytes, ci->i_max_files);
288 static ssize_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
289 char *val, size_t size)
291 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_bytes);
294 static ssize_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
295 char *val, size_t size)
297 return ceph_fmt_xattr(val, size, "%llu", ci->i_max_files);
301 static bool ceph_vxattrcb_snap_btime_exists(struct ceph_inode_info *ci)
303 return (ci->i_snap_btime.tv_sec != 0 || ci->i_snap_btime.tv_nsec != 0);
306 static ssize_t ceph_vxattrcb_snap_btime(struct ceph_inode_info *ci, char *val,
309 return ceph_fmt_xattr(val, size, "%lld.%09ld", ci->i_snap_btime.tv_sec,
310 ci->i_snap_btime.tv_nsec);
313 static ssize_t ceph_vxattrcb_cluster_fsid(struct ceph_inode_info *ci,
314 char *val, size_t size)
316 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
318 return ceph_fmt_xattr(val, size, "%pU", &fsc->client->fsid);
321 static ssize_t ceph_vxattrcb_client_id(struct ceph_inode_info *ci,
322 char *val, size_t size)
324 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
326 return ceph_fmt_xattr(val, size, "client%lld",
327 ceph_client_gid(fsc->client));
330 static ssize_t ceph_vxattrcb_caps(struct ceph_inode_info *ci, char *val,
335 spin_lock(&ci->i_ceph_lock);
336 issued = __ceph_caps_issued(ci, NULL);
337 spin_unlock(&ci->i_ceph_lock);
339 return ceph_fmt_xattr(val, size, "%s/0x%x",
340 ceph_cap_string(issued), issued);
343 static ssize_t ceph_vxattrcb_auth_mds(struct ceph_inode_info *ci,
344 char *val, size_t size)
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);
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
359 #define XATTR_NAME_CEPH(_type, _name, _flags) \
361 .name = CEPH_XATTR_NAME(_type, _name), \
362 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
363 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
365 .flags = (VXATTR_FLAG_READONLY | _flags), \
367 #define XATTR_RSTAT_FIELD(_type, _name) \
368 XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
369 #define XATTR_RSTAT_FIELD_UPDATABLE(_type, _name) \
371 .name = CEPH_XATTR_NAME(_type, _name), \
372 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
373 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
375 .flags = VXATTR_FLAG_RSTAT, \
377 #define XATTR_LAYOUT_FIELD(_type, _name, _field) \
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, \
385 #define XATTR_QUOTA_FIELD(_type, _name) \
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, \
394 static struct ceph_vxattr ceph_dir_vxattrs[] = {
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,
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),
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,
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,
430 XATTR_QUOTA_FIELD(quota, max_bytes),
431 XATTR_QUOTA_FIELD(quota, max_files),
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,
441 .name_size = sizeof("ceph.caps"),
442 .getxattr_cb = ceph_vxattrcb_caps,
444 .flags = VXATTR_FLAG_HIDDEN,
446 { .name = NULL, 0 } /* Required table terminator */
451 static struct ceph_vxattr ceph_file_vxattrs[] = {
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,
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),
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,
473 .name_size = sizeof("ceph.caps"),
474 .getxattr_cb = ceph_vxattrcb_caps,
476 .flags = VXATTR_FLAG_HIDDEN,
478 { .name = NULL, 0 } /* Required table terminator */
481 static struct ceph_vxattr ceph_common_vxattrs[] = {
483 .name = "ceph.cluster_fsid",
484 .name_size = sizeof("ceph.cluster_fsid"),
485 .getxattr_cb = ceph_vxattrcb_cluster_fsid,
487 .flags = VXATTR_FLAG_READONLY,
490 .name = "ceph.client_id",
491 .name_size = sizeof("ceph.client_id"),
492 .getxattr_cb = ceph_vxattrcb_client_id,
494 .flags = VXATTR_FLAG_READONLY,
497 .name = "ceph.auth_mds",
498 .name_size = sizeof("ceph.auth_mds"),
499 .getxattr_cb = ceph_vxattrcb_auth_mds,
501 .flags = VXATTR_FLAG_READONLY,
503 { .name = NULL, 0 } /* Required table terminator */
506 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
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;
515 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
518 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
521 while (vxattr->name) {
522 if (!strcmp(vxattr->name, name))
528 vxattr = ceph_common_vxattrs;
529 while (vxattr->name) {
530 if (!strcmp(vxattr->name, name))
538 static 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)
545 struct rb_node *parent = NULL;
546 struct ceph_inode_xattr *xattr = NULL;
550 p = &ci->i_xattrs.index.rb_node;
553 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
554 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
560 if (name_len == xattr->name_len)
562 else if (name_len < xattr->name_len)
573 if (xattr && (flags & XATTR_CREATE))
575 else if (!xattr && (flags & XATTR_REPLACE))
583 if (update_xattr < 0) {
585 __remove_xattr(ci, xattr);
596 xattr->name_len = name_len;
597 xattr->should_free_name = update_xattr;
599 ci->i_xattrs.count++;
600 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
604 if (xattr->should_free_val)
611 ci->i_xattrs.names_size -= xattr->name_len;
612 ci->i_xattrs.vals_size -= xattr->val_len;
614 ci->i_xattrs.names_size += name_len;
615 ci->i_xattrs.vals_size += val_len;
621 xattr->val_len = val_len;
622 xattr->dirty = update_xattr;
623 xattr->should_free_val = (val && update_xattr);
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);
631 dout("__set_xattr_val added %llx.%llx xattr %p %.*s=%.*s\n",
632 ceph_vinop(&ci->vfs_inode), xattr, name_len, name, val_len, val);
637 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
641 struct rb_node *parent = NULL;
642 struct ceph_inode_xattr *xattr = NULL;
643 int name_len = strlen(name);
646 p = &ci->i_xattrs.index.rb_node;
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)
658 dout("__get_xattr %s: found %.*s\n", name,
659 xattr->val_len, xattr->val);
664 dout("__get_xattr %s: not found\n", name);
669 static void __free_xattr(struct ceph_inode_xattr *xattr)
673 if (xattr->should_free_name)
675 if (xattr->should_free_val)
681 static int __remove_xattr(struct ceph_inode_info *ci,
682 struct ceph_inode_xattr *xattr)
687 rb_erase(&xattr->node, &ci->i_xattrs.index);
689 if (xattr->should_free_name)
691 if (xattr->should_free_val)
694 ci->i_xattrs.names_size -= xattr->name_len;
695 ci->i_xattrs.vals_size -= xattr->val_len;
696 ci->i_xattrs.count--;
702 static char *__copy_xattr_names(struct ceph_inode_info *ci,
706 struct ceph_inode_xattr *xattr = NULL;
708 p = rb_first(&ci->i_xattrs.index);
709 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
712 xattr = rb_entry(p, struct ceph_inode_xattr, node);
713 memcpy(dest, xattr->name, xattr->name_len);
714 dest[xattr->name_len] = '\0';
716 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
717 xattr->name_len, ci->i_xattrs.names_size);
719 dest += xattr->name_len + 1;
726 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
728 struct rb_node *p, *tmp;
729 struct ceph_inode_xattr *xattr = NULL;
731 p = rb_first(&ci->i_xattrs.index);
733 dout("__ceph_destroy_xattrs p=%p\n", p);
736 xattr = rb_entry(p, struct ceph_inode_xattr, node);
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);
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;
753 static int __build_xattrs(struct inode *inode)
754 __releases(ci->i_ceph_lock)
755 __acquires(ci->i_ceph_lock)
761 const char *name, *val;
762 struct ceph_inode_info *ci = ceph_inode(inode);
764 struct ceph_inode_xattr **xattrs = NULL;
768 dout("__build_xattrs() len=%d\n",
769 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
771 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
772 return 0; /* already built */
774 __ceph_destroy_xattrs(ci);
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);
785 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
791 for (i = 0; i < numattr; i++) {
792 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
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++)
809 ceph_decode_32_safe(&p, end, len, bad);
813 ceph_decode_32_safe(&p, end, len, bad);
817 err = __set_xattr(ci, name, namelen, val, len,
818 0, 0, &xattrs[numattr]);
825 ci->i_xattrs.index_version = ci->i_xattrs.version;
826 ci->i_xattrs.dirty = false;
830 spin_lock(&ci->i_ceph_lock);
833 for (i = 0; i < numattr; i++)
837 ci->i_xattrs.names_size = 0;
841 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
845 * 4 bytes for the length, and additional 4 bytes per each xattr name,
846 * 4 bytes per each value
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);
856 size += 4 + 4 + name_size + val_size;
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
867 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
870 struct ceph_inode_xattr *xattr = NULL;
871 struct ceph_buffer *old_blob = NULL;
874 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
875 if (ci->i_xattrs.dirty) {
876 int need = __get_required_blob_size(ci, 0, 0);
878 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
880 p = rb_first(&ci->i_xattrs.index);
881 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
883 ceph_encode_32(&dest, ci->i_xattrs.count);
885 xattr = rb_entry(p, struct ceph_inode_xattr, node);
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;
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;
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++;
912 static inline int __get_request_mask(struct inode *in) {
913 struct ceph_mds_request *req = current->journal_info;
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);
929 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
932 struct ceph_inode_info *ci = ceph_inode(inode);
933 struct ceph_inode_xattr *xattr;
934 struct ceph_vxattr *vxattr;
938 if (strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
939 goto handle_non_vxattrs;
941 /* let's see if a virtual xattr was requested */
942 vxattr = ceph_match_vxattr(inode, name);
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);
953 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
954 err = vxattr->getxattr_cb(ci, value, size);
955 if (size && size < err)
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)
967 req_mask = __get_request_mask(inode);
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);
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);
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);
985 /* get xattrs from mds (if we don't already have them) */
986 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
989 spin_lock(&ci->i_ceph_lock);
992 err = __build_xattrs(inode);
996 err = -ENODATA; /* == ENOATTR */
997 xattr = __get_xattr(ci, name);
1002 if (size && size < xattr->val_len)
1005 err = xattr->val_len;
1009 memcpy(value, xattr->val, xattr->val_len);
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;
1016 spin_unlock(&ci->i_ceph_lock);
1020 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
1022 struct inode *inode = d_inode(dentry);
1023 struct ceph_inode_info *ci = ceph_inode(inode);
1024 bool len_only = (size == 0);
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);
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);
1038 spin_lock(&ci->i_ceph_lock);
1041 err = __build_xattrs(inode);
1045 /* add 1 byte for each xattr due to the null termination */
1046 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
1048 if (namelen > size) {
1052 names = __copy_xattr_names(ci, names);
1057 spin_unlock(&ci->i_ceph_lock);
1061 static int ceph_sync_setxattr(struct inode *inode, const char *name,
1062 const char *value, size_t size, int flags)
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;
1074 /* copy value into pagelist */
1075 pagelist = ceph_pagelist_alloc(GFP_NOFS);
1079 err = ceph_pagelist_append(pagelist, value, size);
1082 } else if (!value) {
1083 if (flags & CEPH_XATTR_REPLACE)
1084 op = CEPH_MDS_OP_RMXATTR;
1086 flags |= CEPH_XATTR_REMOVE;
1089 dout("setxattr value=%.*s\n", (int)size, value);
1092 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1098 req->r_path2 = kstrdup(name, GFP_NOFS);
1099 if (!req->r_path2) {
1100 ceph_mdsc_put_request(req);
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;
1113 req->r_inode = inode;
1115 req->r_num_caps = 1;
1116 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
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);
1125 ceph_pagelist_release(pagelist);
1129 int __ceph_setxattr(struct inode *inode, const char *name,
1130 const void *value, size_t size, int flags)
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;
1140 int name_len = strlen(name);
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;
1149 if (ceph_snap(inode) != CEPH_NOSNAP)
1152 vxattr = ceph_match_vxattr(inode, name);
1154 if (vxattr->flags & VXATTR_FLAG_READONLY)
1156 if (value && !strncmp(vxattr->name, "ceph.quota", 10))
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;
1164 /* preallocate memory for xattr name, value, index node */
1166 newname = kmemdup(name, name_len + 1, GFP_NOFS);
1171 newval = kmemdup(value, val_len, GFP_NOFS);
1176 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1180 prealloc_cf = ceph_alloc_cap_flush();
1184 spin_lock(&ci->i_ceph_lock);
1186 issued = __ceph_caps_issued(ci, NULL);
1187 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1190 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1191 lock_snap_rwsem = true;
1192 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1193 spin_unlock(&ci->i_ceph_lock);
1194 down_read(&mdsc->snap_rwsem);
1195 spin_lock(&ci->i_ceph_lock);
1200 dout("setxattr %p name '%s' issued %s\n", inode, name,
1201 ceph_cap_string(issued));
1202 __build_xattrs(inode);
1204 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1206 if (!ci->i_xattrs.prealloc_blob ||
1207 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1208 struct ceph_buffer *blob;
1210 spin_unlock(&ci->i_ceph_lock);
1211 ceph_buffer_put(old_blob); /* Shouldn't be required */
1212 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1213 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1215 goto do_sync_unlocked;
1216 spin_lock(&ci->i_ceph_lock);
1217 /* prealloc_blob can't be released while holding i_ceph_lock */
1218 if (ci->i_xattrs.prealloc_blob)
1219 old_blob = ci->i_xattrs.prealloc_blob;
1220 ci->i_xattrs.prealloc_blob = blob;
1224 err = __set_xattr(ci, newname, name_len, newval, val_len,
1225 flags, value ? 1 : -1, &xattr);
1228 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1230 ci->i_xattrs.dirty = true;
1231 inode->i_ctime = current_time(inode);
1234 spin_unlock(&ci->i_ceph_lock);
1235 ceph_buffer_put(old_blob);
1236 if (lock_snap_rwsem)
1237 up_read(&mdsc->snap_rwsem);
1239 __mark_inode_dirty(inode, dirty);
1240 ceph_free_cap_flush(prealloc_cf);
1244 spin_unlock(&ci->i_ceph_lock);
1246 if (lock_snap_rwsem)
1247 up_read(&mdsc->snap_rwsem);
1249 /* security module set xattr while filling trace */
1250 if (current->journal_info) {
1251 pr_warn_ratelimited("sync setxattr %p "
1252 "during filling trace\n", inode);
1255 err = ceph_sync_setxattr(inode, name, value, size, flags);
1256 if (err >= 0 && check_realm) {
1257 /* check if snaprealm was created for quota inode */
1258 spin_lock(&ci->i_ceph_lock);
1259 if ((ci->i_max_files || ci->i_max_bytes) &&
1260 !(ci->i_snap_realm &&
1261 ci->i_snap_realm->ino == ci->i_vino.ino))
1263 spin_unlock(&ci->i_ceph_lock);
1267 ceph_free_cap_flush(prealloc_cf);
1274 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1275 struct dentry *dentry, struct inode *inode,
1276 const char *name, void *value, size_t size)
1278 if (!ceph_is_valid_xattr(name))
1280 return __ceph_getxattr(inode, name, value, size);
1283 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1284 struct user_namespace *mnt_userns,
1285 struct dentry *unused, struct inode *inode,
1286 const char *name, const void *value,
1287 size_t size, int flags)
1289 if (!ceph_is_valid_xattr(name))
1291 return __ceph_setxattr(inode, name, value, size, flags);
1294 static const struct xattr_handler ceph_other_xattr_handler = {
1295 .prefix = "", /* match any name => handlers called with full name */
1296 .get = ceph_get_xattr_handler,
1297 .set = ceph_set_xattr_handler,
1300 #ifdef CONFIG_SECURITY
1301 bool ceph_security_xattr_wanted(struct inode *in)
1303 return in->i_security != NULL;
1306 bool ceph_security_xattr_deadlock(struct inode *in)
1308 struct ceph_inode_info *ci;
1310 if (!in->i_security)
1312 ci = ceph_inode(in);
1313 spin_lock(&ci->i_ceph_lock);
1314 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1315 !(ci->i_xattrs.version > 0 &&
1316 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1317 spin_unlock(&ci->i_ceph_lock);
1321 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1322 int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1323 struct ceph_acl_sec_ctx *as_ctx)
1325 struct ceph_pagelist *pagelist = as_ctx->pagelist;
1330 err = security_dentry_init_security(dentry, mode, &dentry->d_name,
1331 &name, &as_ctx->sec_ctx,
1332 &as_ctx->sec_ctxlen);
1334 WARN_ON_ONCE(err != -EOPNOTSUPP);
1335 err = 0; /* do nothing */
1341 pagelist = ceph_pagelist_alloc(GFP_KERNEL);
1344 err = ceph_pagelist_reserve(pagelist, PAGE_SIZE);
1347 ceph_pagelist_encode_32(pagelist, 1);
1351 * FIXME: Make security_dentry_init_security() generic. Currently
1352 * It only supports single security module and only selinux has
1353 * dentry_init_security hook.
1355 name_len = strlen(name);
1356 err = ceph_pagelist_reserve(pagelist,
1357 4 * 2 + name_len + as_ctx->sec_ctxlen);
1361 if (as_ctx->pagelist) {
1362 /* update count of KV pairs */
1363 BUG_ON(pagelist->length <= sizeof(__le32));
1364 if (list_is_singular(&pagelist->head)) {
1365 le32_add_cpu((__le32*)pagelist->mapped_tail, 1);
1367 struct page *page = list_first_entry(&pagelist->head,
1369 void *addr = kmap_atomic(page);
1370 le32_add_cpu((__le32*)addr, 1);
1371 kunmap_atomic(addr);
1374 as_ctx->pagelist = pagelist;
1377 ceph_pagelist_encode_32(pagelist, name_len);
1378 ceph_pagelist_append(pagelist, name, name_len);
1380 ceph_pagelist_encode_32(pagelist, as_ctx->sec_ctxlen);
1381 ceph_pagelist_append(pagelist, as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1385 if (pagelist && !as_ctx->pagelist)
1386 ceph_pagelist_release(pagelist);
1389 #endif /* CONFIG_CEPH_FS_SECURITY_LABEL */
1390 #endif /* CONFIG_SECURITY */
1392 void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx)
1394 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1395 posix_acl_release(as_ctx->acl);
1396 posix_acl_release(as_ctx->default_acl);
1398 #ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1399 security_release_secctx(as_ctx->sec_ctx, as_ctx->sec_ctxlen);
1401 if (as_ctx->pagelist)
1402 ceph_pagelist_release(as_ctx->pagelist);
1406 * List of handlers for synthetic system.* attributes. Other
1407 * attributes are handled directly.
1409 const struct xattr_handler *ceph_xattr_handlers[] = {
1410 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1411 &posix_acl_access_xattr_handler,
1412 &posix_acl_default_xattr_handler,
1414 &ceph_other_xattr_handler,