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/posix_acl_xattr.h>
12 #include <linux/slab.h>
14 #define XATTR_CEPH_PREFIX "ceph."
15 #define XATTR_CEPH_PREFIX_LEN (sizeof (XATTR_CEPH_PREFIX) - 1)
17 static int __remove_xattr(struct ceph_inode_info *ci,
18 struct ceph_inode_xattr *xattr);
20 static const struct xattr_handler ceph_other_xattr_handler;
23 * List of handlers for synthetic system.* attributes. Other
24 * attributes are handled directly.
26 const 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,
31 &ceph_other_xattr_handler,
35 static bool ceph_is_valid_xattr(const char *name)
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);
45 * These define virtual xattrs exposing the recursive directory
46 * statistics and layout metadata.
50 size_t name_size; /* strlen(name) + 1 (for '\0') */
51 size_t (*getxattr_cb)(struct ceph_inode_info *ci, char *val,
53 bool (*exists_cb)(struct ceph_inode_info *ci);
57 #define VXATTR_FLAG_READONLY (1<<0)
58 #define VXATTR_FLAG_HIDDEN (1<<1)
59 #define VXATTR_FLAG_RSTAT (1<<2)
63 static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
65 struct ceph_file_layout *fl = &ci->i_layout;
66 return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
67 fl->object_size > 0 || fl->pool_id >= 0 ||
68 rcu_dereference_raw(fl->pool_ns) != NULL);
71 static size_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
74 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
75 struct ceph_osd_client *osdc = &fsc->client->osdc;
76 struct ceph_string *pool_ns;
77 s64 pool = ci->i_layout.pool_id;
78 const char *pool_name;
79 const char *ns_field = " pool_namespace=";
81 size_t len, total_len = 0;
84 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
86 dout("ceph_vxattrcb_layout %p\n", &ci->vfs_inode);
87 down_read(&osdc->lock);
88 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
90 len = snprintf(buf, sizeof(buf),
91 "stripe_unit=%u stripe_count=%u object_size=%u pool=",
92 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
93 ci->i_layout.object_size);
94 total_len = len + strlen(pool_name);
96 len = snprintf(buf, sizeof(buf),
97 "stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
98 ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
99 ci->i_layout.object_size, (unsigned long long)pool);
104 total_len += strlen(ns_field) + pool_ns->len;
107 if (size >= total_len) {
108 memcpy(val, buf, len);
111 len = strlen(pool_name);
112 memcpy(val + ret, pool_name, len);
116 len = strlen(ns_field);
117 memcpy(val + ret, ns_field, len);
119 memcpy(val + ret, pool_ns->str, pool_ns->len);
123 up_read(&osdc->lock);
124 ceph_put_string(pool_ns);
128 static size_t ceph_vxattrcb_layout_stripe_unit(struct ceph_inode_info *ci,
129 char *val, size_t size)
131 return snprintf(val, size, "%u", ci->i_layout.stripe_unit);
134 static size_t ceph_vxattrcb_layout_stripe_count(struct ceph_inode_info *ci,
135 char *val, size_t size)
137 return snprintf(val, size, "%u", ci->i_layout.stripe_count);
140 static size_t ceph_vxattrcb_layout_object_size(struct ceph_inode_info *ci,
141 char *val, size_t size)
143 return snprintf(val, size, "%u", ci->i_layout.object_size);
146 static size_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
147 char *val, size_t size)
150 struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
151 struct ceph_osd_client *osdc = &fsc->client->osdc;
152 s64 pool = ci->i_layout.pool_id;
153 const char *pool_name;
155 down_read(&osdc->lock);
156 pool_name = ceph_pg_pool_name_by_id(osdc->osdmap, pool);
158 ret = snprintf(val, size, "%s", pool_name);
160 ret = snprintf(val, size, "%lld", (unsigned long long)pool);
161 up_read(&osdc->lock);
165 static size_t ceph_vxattrcb_layout_pool_namespace(struct ceph_inode_info *ci,
166 char *val, size_t size)
169 struct ceph_string *ns = ceph_try_get_string(ci->i_layout.pool_ns);
171 ret = snprintf(val, size, "%.*s", (int)ns->len, ns->str);
179 static size_t ceph_vxattrcb_dir_entries(struct ceph_inode_info *ci, char *val,
182 return snprintf(val, size, "%lld", ci->i_files + ci->i_subdirs);
185 static size_t ceph_vxattrcb_dir_files(struct ceph_inode_info *ci, char *val,
188 return snprintf(val, size, "%lld", ci->i_files);
191 static size_t ceph_vxattrcb_dir_subdirs(struct ceph_inode_info *ci, char *val,
194 return snprintf(val, size, "%lld", ci->i_subdirs);
197 static size_t ceph_vxattrcb_dir_rentries(struct ceph_inode_info *ci, char *val,
200 return snprintf(val, size, "%lld", ci->i_rfiles + ci->i_rsubdirs);
203 static size_t ceph_vxattrcb_dir_rfiles(struct ceph_inode_info *ci, char *val,
206 return snprintf(val, size, "%lld", ci->i_rfiles);
209 static size_t ceph_vxattrcb_dir_rsubdirs(struct ceph_inode_info *ci, char *val,
212 return snprintf(val, size, "%lld", ci->i_rsubdirs);
215 static size_t ceph_vxattrcb_dir_rbytes(struct ceph_inode_info *ci, char *val,
218 return snprintf(val, size, "%lld", ci->i_rbytes);
221 static size_t ceph_vxattrcb_dir_rctime(struct ceph_inode_info *ci, char *val,
224 return snprintf(val, size, "%lld.%09ld", ci->i_rctime.tv_sec,
225 ci->i_rctime.tv_nsec);
230 static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
233 spin_lock(&ci->i_ceph_lock);
234 if ((ci->i_max_files || ci->i_max_bytes) &&
235 ci->i_vino.snap == CEPH_NOSNAP &&
237 ci->i_snap_realm->ino == ci->i_vino.ino)
239 spin_unlock(&ci->i_ceph_lock);
243 static size_t ceph_vxattrcb_quota(struct ceph_inode_info *ci, char *val,
246 return snprintf(val, size, "max_bytes=%llu max_files=%llu",
247 ci->i_max_bytes, ci->i_max_files);
250 static size_t ceph_vxattrcb_quota_max_bytes(struct ceph_inode_info *ci,
251 char *val, size_t size)
253 return snprintf(val, size, "%llu", ci->i_max_bytes);
256 static size_t ceph_vxattrcb_quota_max_files(struct ceph_inode_info *ci,
257 char *val, size_t size)
259 return snprintf(val, size, "%llu", ci->i_max_files);
262 #define CEPH_XATTR_NAME(_type, _name) XATTR_CEPH_PREFIX #_type "." #_name
263 #define CEPH_XATTR_NAME2(_type, _name, _name2) \
264 XATTR_CEPH_PREFIX #_type "." #_name "." #_name2
266 #define XATTR_NAME_CEPH(_type, _name, _flags) \
268 .name = CEPH_XATTR_NAME(_type, _name), \
269 .name_size = sizeof (CEPH_XATTR_NAME(_type, _name)), \
270 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
272 .flags = (VXATTR_FLAG_READONLY | _flags), \
274 #define XATTR_RSTAT_FIELD(_type, _name) \
275 XATTR_NAME_CEPH(_type, _name, VXATTR_FLAG_RSTAT)
276 #define XATTR_LAYOUT_FIELD(_type, _name, _field) \
278 .name = CEPH_XATTR_NAME2(_type, _name, _field), \
279 .name_size = sizeof (CEPH_XATTR_NAME2(_type, _name, _field)), \
280 .getxattr_cb = ceph_vxattrcb_ ## _name ## _ ## _field, \
281 .exists_cb = ceph_vxattrcb_layout_exists, \
282 .flags = VXATTR_FLAG_HIDDEN, \
284 #define XATTR_QUOTA_FIELD(_type, _name) \
286 .name = CEPH_XATTR_NAME(_type, _name), \
287 .name_size = sizeof(CEPH_XATTR_NAME(_type, _name)), \
288 .getxattr_cb = ceph_vxattrcb_ ## _type ## _ ## _name, \
289 .exists_cb = ceph_vxattrcb_quota_exists, \
290 .flags = VXATTR_FLAG_HIDDEN, \
293 static struct ceph_vxattr ceph_dir_vxattrs[] = {
295 .name = "ceph.dir.layout",
296 .name_size = sizeof("ceph.dir.layout"),
297 .getxattr_cb = ceph_vxattrcb_layout,
298 .exists_cb = ceph_vxattrcb_layout_exists,
299 .flags = VXATTR_FLAG_HIDDEN,
301 XATTR_LAYOUT_FIELD(dir, layout, stripe_unit),
302 XATTR_LAYOUT_FIELD(dir, layout, stripe_count),
303 XATTR_LAYOUT_FIELD(dir, layout, object_size),
304 XATTR_LAYOUT_FIELD(dir, layout, pool),
305 XATTR_LAYOUT_FIELD(dir, layout, pool_namespace),
306 XATTR_NAME_CEPH(dir, entries, 0),
307 XATTR_NAME_CEPH(dir, files, 0),
308 XATTR_NAME_CEPH(dir, subdirs, 0),
309 XATTR_RSTAT_FIELD(dir, rentries),
310 XATTR_RSTAT_FIELD(dir, rfiles),
311 XATTR_RSTAT_FIELD(dir, rsubdirs),
312 XATTR_RSTAT_FIELD(dir, rbytes),
313 XATTR_RSTAT_FIELD(dir, rctime),
315 .name = "ceph.quota",
316 .name_size = sizeof("ceph.quota"),
317 .getxattr_cb = ceph_vxattrcb_quota,
318 .exists_cb = ceph_vxattrcb_quota_exists,
319 .flags = VXATTR_FLAG_HIDDEN,
321 XATTR_QUOTA_FIELD(quota, max_bytes),
322 XATTR_QUOTA_FIELD(quota, max_files),
323 { .name = NULL, 0 } /* Required table terminator */
325 static size_t ceph_dir_vxattrs_name_size; /* total size of all names */
329 static struct ceph_vxattr ceph_file_vxattrs[] = {
331 .name = "ceph.file.layout",
332 .name_size = sizeof("ceph.file.layout"),
333 .getxattr_cb = ceph_vxattrcb_layout,
334 .exists_cb = ceph_vxattrcb_layout_exists,
335 .flags = VXATTR_FLAG_HIDDEN,
337 XATTR_LAYOUT_FIELD(file, layout, stripe_unit),
338 XATTR_LAYOUT_FIELD(file, layout, stripe_count),
339 XATTR_LAYOUT_FIELD(file, layout, object_size),
340 XATTR_LAYOUT_FIELD(file, layout, pool),
341 XATTR_LAYOUT_FIELD(file, layout, pool_namespace),
342 { .name = NULL, 0 } /* Required table terminator */
344 static size_t ceph_file_vxattrs_name_size; /* total size of all names */
346 static struct ceph_vxattr *ceph_inode_vxattrs(struct inode *inode)
348 if (S_ISDIR(inode->i_mode))
349 return ceph_dir_vxattrs;
350 else if (S_ISREG(inode->i_mode))
351 return ceph_file_vxattrs;
355 static size_t ceph_vxattrs_name_size(struct ceph_vxattr *vxattrs)
357 if (vxattrs == ceph_dir_vxattrs)
358 return ceph_dir_vxattrs_name_size;
359 if (vxattrs == ceph_file_vxattrs)
360 return ceph_file_vxattrs_name_size;
366 * Compute the aggregate size (including terminating '\0') of all
367 * virtual extended attribute names in the given vxattr table.
369 static size_t __init vxattrs_name_size(struct ceph_vxattr *vxattrs)
371 struct ceph_vxattr *vxattr;
374 for (vxattr = vxattrs; vxattr->name; vxattr++) {
375 if (!(vxattr->flags & VXATTR_FLAG_HIDDEN))
376 size += vxattr->name_size;
382 /* Routines called at initialization and exit time */
384 void __init ceph_xattr_init(void)
386 ceph_dir_vxattrs_name_size = vxattrs_name_size(ceph_dir_vxattrs);
387 ceph_file_vxattrs_name_size = vxattrs_name_size(ceph_file_vxattrs);
390 void ceph_xattr_exit(void)
392 ceph_dir_vxattrs_name_size = 0;
393 ceph_file_vxattrs_name_size = 0;
396 static struct ceph_vxattr *ceph_match_vxattr(struct inode *inode,
399 struct ceph_vxattr *vxattr = ceph_inode_vxattrs(inode);
402 while (vxattr->name) {
403 if (!strcmp(vxattr->name, name))
412 static int __set_xattr(struct ceph_inode_info *ci,
413 const char *name, int name_len,
414 const char *val, int val_len,
415 int flags, int update_xattr,
416 struct ceph_inode_xattr **newxattr)
419 struct rb_node *parent = NULL;
420 struct ceph_inode_xattr *xattr = NULL;
424 p = &ci->i_xattrs.index.rb_node;
427 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
428 c = strncmp(name, xattr->name, min(name_len, xattr->name_len));
434 if (name_len == xattr->name_len)
436 else if (name_len < xattr->name_len)
447 if (xattr && (flags & XATTR_CREATE))
449 else if (!xattr && (flags & XATTR_REPLACE))
457 if (update_xattr < 0) {
459 __remove_xattr(ci, xattr);
470 xattr->name_len = name_len;
471 xattr->should_free_name = update_xattr;
473 ci->i_xattrs.count++;
474 dout("__set_xattr count=%d\n", ci->i_xattrs.count);
478 if (xattr->should_free_val)
479 kfree((void *)xattr->val);
485 ci->i_xattrs.names_size -= xattr->name_len;
486 ci->i_xattrs.vals_size -= xattr->val_len;
488 ci->i_xattrs.names_size += name_len;
489 ci->i_xattrs.vals_size += val_len;
495 xattr->val_len = val_len;
496 xattr->dirty = update_xattr;
497 xattr->should_free_val = (val && update_xattr);
500 rb_link_node(&xattr->node, parent, p);
501 rb_insert_color(&xattr->node, &ci->i_xattrs.index);
502 dout("__set_xattr_val p=%p\n", p);
505 dout("__set_xattr_val added %llx.%llx xattr %p %s=%.*s\n",
506 ceph_vinop(&ci->vfs_inode), xattr, name, val_len, val);
511 static struct ceph_inode_xattr *__get_xattr(struct ceph_inode_info *ci,
515 struct rb_node *parent = NULL;
516 struct ceph_inode_xattr *xattr = NULL;
517 int name_len = strlen(name);
520 p = &ci->i_xattrs.index.rb_node;
523 xattr = rb_entry(parent, struct ceph_inode_xattr, node);
524 c = strncmp(name, xattr->name, xattr->name_len);
525 if (c == 0 && name_len > xattr->name_len)
532 dout("__get_xattr %s: found %.*s\n", name,
533 xattr->val_len, xattr->val);
538 dout("__get_xattr %s: not found\n", name);
543 static void __free_xattr(struct ceph_inode_xattr *xattr)
547 if (xattr->should_free_name)
548 kfree((void *)xattr->name);
549 if (xattr->should_free_val)
550 kfree((void *)xattr->val);
555 static int __remove_xattr(struct ceph_inode_info *ci,
556 struct ceph_inode_xattr *xattr)
561 rb_erase(&xattr->node, &ci->i_xattrs.index);
563 if (xattr->should_free_name)
564 kfree((void *)xattr->name);
565 if (xattr->should_free_val)
566 kfree((void *)xattr->val);
568 ci->i_xattrs.names_size -= xattr->name_len;
569 ci->i_xattrs.vals_size -= xattr->val_len;
570 ci->i_xattrs.count--;
576 static char *__copy_xattr_names(struct ceph_inode_info *ci,
580 struct ceph_inode_xattr *xattr = NULL;
582 p = rb_first(&ci->i_xattrs.index);
583 dout("__copy_xattr_names count=%d\n", ci->i_xattrs.count);
586 xattr = rb_entry(p, struct ceph_inode_xattr, node);
587 memcpy(dest, xattr->name, xattr->name_len);
588 dest[xattr->name_len] = '\0';
590 dout("dest=%s %p (%s) (%d/%d)\n", dest, xattr, xattr->name,
591 xattr->name_len, ci->i_xattrs.names_size);
593 dest += xattr->name_len + 1;
600 void __ceph_destroy_xattrs(struct ceph_inode_info *ci)
602 struct rb_node *p, *tmp;
603 struct ceph_inode_xattr *xattr = NULL;
605 p = rb_first(&ci->i_xattrs.index);
607 dout("__ceph_destroy_xattrs p=%p\n", p);
610 xattr = rb_entry(p, struct ceph_inode_xattr, node);
613 dout("__ceph_destroy_xattrs next p=%p (%.*s)\n", p,
614 xattr->name_len, xattr->name);
615 rb_erase(tmp, &ci->i_xattrs.index);
620 ci->i_xattrs.names_size = 0;
621 ci->i_xattrs.vals_size = 0;
622 ci->i_xattrs.index_version = 0;
623 ci->i_xattrs.count = 0;
624 ci->i_xattrs.index = RB_ROOT;
627 static int __build_xattrs(struct inode *inode)
628 __releases(ci->i_ceph_lock)
629 __acquires(ci->i_ceph_lock)
635 const char *name, *val;
636 struct ceph_inode_info *ci = ceph_inode(inode);
638 struct ceph_inode_xattr **xattrs = NULL;
642 dout("__build_xattrs() len=%d\n",
643 ci->i_xattrs.blob ? (int)ci->i_xattrs.blob->vec.iov_len : 0);
645 if (ci->i_xattrs.index_version >= ci->i_xattrs.version)
646 return 0; /* already built */
648 __ceph_destroy_xattrs(ci);
651 /* updated internal xattr rb tree */
652 if (ci->i_xattrs.blob && ci->i_xattrs.blob->vec.iov_len > 4) {
653 p = ci->i_xattrs.blob->vec.iov_base;
654 end = p + ci->i_xattrs.blob->vec.iov_len;
655 ceph_decode_32_safe(&p, end, numattr, bad);
656 xattr_version = ci->i_xattrs.version;
657 spin_unlock(&ci->i_ceph_lock);
659 xattrs = kcalloc(numattr, sizeof(struct ceph_inode_xattr *),
665 for (i = 0; i < numattr; i++) {
666 xattrs[i] = kmalloc(sizeof(struct ceph_inode_xattr),
672 spin_lock(&ci->i_ceph_lock);
673 if (ci->i_xattrs.version != xattr_version) {
674 /* lost a race, retry */
675 for (i = 0; i < numattr; i++)
683 ceph_decode_32_safe(&p, end, len, bad);
687 ceph_decode_32_safe(&p, end, len, bad);
691 err = __set_xattr(ci, name, namelen, val, len,
692 0, 0, &xattrs[numattr]);
699 ci->i_xattrs.index_version = ci->i_xattrs.version;
700 ci->i_xattrs.dirty = false;
704 spin_lock(&ci->i_ceph_lock);
707 for (i = 0; i < numattr; i++)
711 ci->i_xattrs.names_size = 0;
715 static int __get_required_blob_size(struct ceph_inode_info *ci, int name_size,
719 * 4 bytes for the length, and additional 4 bytes per each xattr name,
720 * 4 bytes per each value
722 int size = 4 + ci->i_xattrs.count*(4 + 4) +
723 ci->i_xattrs.names_size +
724 ci->i_xattrs.vals_size;
725 dout("__get_required_blob_size c=%d names.size=%d vals.size=%d\n",
726 ci->i_xattrs.count, ci->i_xattrs.names_size,
727 ci->i_xattrs.vals_size);
730 size += 4 + 4 + name_size + val_size;
736 * If there are dirty xattrs, reencode xattrs into the prealloc_blob
737 * and swap into place. It returns the old i_xattrs.blob (or NULL) so
738 * that it can be freed by the caller as the i_ceph_lock is likely to be
741 struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci)
744 struct ceph_inode_xattr *xattr = NULL;
745 struct ceph_buffer *old_blob = NULL;
748 dout("__build_xattrs_blob %p\n", &ci->vfs_inode);
749 if (ci->i_xattrs.dirty) {
750 int need = __get_required_blob_size(ci, 0, 0);
752 BUG_ON(need > ci->i_xattrs.prealloc_blob->alloc_len);
754 p = rb_first(&ci->i_xattrs.index);
755 dest = ci->i_xattrs.prealloc_blob->vec.iov_base;
757 ceph_encode_32(&dest, ci->i_xattrs.count);
759 xattr = rb_entry(p, struct ceph_inode_xattr, node);
761 ceph_encode_32(&dest, xattr->name_len);
762 memcpy(dest, xattr->name, xattr->name_len);
763 dest += xattr->name_len;
764 ceph_encode_32(&dest, xattr->val_len);
765 memcpy(dest, xattr->val, xattr->val_len);
766 dest += xattr->val_len;
771 /* adjust buffer len; it may be larger than we need */
772 ci->i_xattrs.prealloc_blob->vec.iov_len =
773 dest - ci->i_xattrs.prealloc_blob->vec.iov_base;
775 if (ci->i_xattrs.blob)
776 old_blob = ci->i_xattrs.blob;
777 ci->i_xattrs.blob = ci->i_xattrs.prealloc_blob;
778 ci->i_xattrs.prealloc_blob = NULL;
779 ci->i_xattrs.dirty = false;
780 ci->i_xattrs.version++;
786 static inline int __get_request_mask(struct inode *in) {
787 struct ceph_mds_request *req = current->journal_info;
789 if (req && req->r_target_inode == in) {
790 if (req->r_op == CEPH_MDS_OP_LOOKUP ||
791 req->r_op == CEPH_MDS_OP_LOOKUPINO ||
792 req->r_op == CEPH_MDS_OP_LOOKUPPARENT ||
793 req->r_op == CEPH_MDS_OP_GETATTR) {
794 mask = le32_to_cpu(req->r_args.getattr.mask);
795 } else if (req->r_op == CEPH_MDS_OP_OPEN ||
796 req->r_op == CEPH_MDS_OP_CREATE) {
797 mask = le32_to_cpu(req->r_args.open.mask);
803 ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,
806 struct ceph_inode_info *ci = ceph_inode(inode);
807 struct ceph_inode_xattr *xattr;
808 struct ceph_vxattr *vxattr = NULL;
812 /* let's see if a virtual xattr was requested */
813 vxattr = ceph_match_vxattr(inode, name);
816 if (vxattr->flags & VXATTR_FLAG_RSTAT)
817 mask |= CEPH_STAT_RSTAT;
818 err = ceph_do_getattr(inode, mask, true);
822 if (!(vxattr->exists_cb && !vxattr->exists_cb(ci))) {
823 err = vxattr->getxattr_cb(ci, value, size);
824 if (size && size < err)
830 req_mask = __get_request_mask(inode);
832 spin_lock(&ci->i_ceph_lock);
833 dout("getxattr %p ver=%lld index_ver=%lld\n", inode,
834 ci->i_xattrs.version, ci->i_xattrs.index_version);
836 if (ci->i_xattrs.version == 0 ||
837 !((req_mask & CEPH_CAP_XATTR_SHARED) ||
838 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1))) {
839 spin_unlock(&ci->i_ceph_lock);
841 /* security module gets xattr while filling trace */
842 if (current->journal_info) {
843 pr_warn_ratelimited("sync getxattr %p "
844 "during filling trace\n", inode);
848 /* get xattrs from mds (if we don't already have them) */
849 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
852 spin_lock(&ci->i_ceph_lock);
855 err = __build_xattrs(inode);
859 err = -ENODATA; /* == ENOATTR */
860 xattr = __get_xattr(ci, name);
865 if (size && size < xattr->val_len)
868 err = xattr->val_len;
872 memcpy(value, xattr->val, xattr->val_len);
874 if (current->journal_info &&
875 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
876 ci->i_ceph_flags |= CEPH_I_SEC_INITED;
878 spin_unlock(&ci->i_ceph_lock);
882 ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)
884 struct inode *inode = d_inode(dentry);
885 struct ceph_inode_info *ci = ceph_inode(inode);
886 struct ceph_vxattr *vxattrs = ceph_inode_vxattrs(inode);
893 spin_lock(&ci->i_ceph_lock);
894 dout("listxattr %p ver=%lld index_ver=%lld\n", inode,
895 ci->i_xattrs.version, ci->i_xattrs.index_version);
897 if (ci->i_xattrs.version == 0 ||
898 !__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 1)) {
899 spin_unlock(&ci->i_ceph_lock);
900 err = ceph_do_getattr(inode, CEPH_STAT_CAP_XATTR, true);
903 spin_lock(&ci->i_ceph_lock);
906 err = __build_xattrs(inode);
910 * Start with virtual dir xattr names (if any) (including
911 * terminating '\0' characters for each).
913 vir_namelen = ceph_vxattrs_name_size(vxattrs);
915 /* adding 1 byte per each variable due to the null termination */
916 namelen = ci->i_xattrs.names_size + ci->i_xattrs.count;
918 if (size && vir_namelen + namelen > size)
921 err = namelen + vir_namelen;
925 names = __copy_xattr_names(ci, names);
927 /* virtual xattr names, too */
930 for (i = 0; vxattrs[i].name; i++) {
931 if (!(vxattrs[i].flags & VXATTR_FLAG_HIDDEN) &&
932 !(vxattrs[i].exists_cb &&
933 !vxattrs[i].exists_cb(ci))) {
934 len = sprintf(names, "%s", vxattrs[i].name);
942 spin_unlock(&ci->i_ceph_lock);
946 static int ceph_sync_setxattr(struct inode *inode, const char *name,
947 const char *value, size_t size, int flags)
949 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
950 struct ceph_inode_info *ci = ceph_inode(inode);
951 struct ceph_mds_request *req;
952 struct ceph_mds_client *mdsc = fsc->mdsc;
953 struct ceph_pagelist *pagelist = NULL;
954 int op = CEPH_MDS_OP_SETXATTR;
958 /* copy value into pagelist */
959 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
963 ceph_pagelist_init(pagelist);
964 err = ceph_pagelist_append(pagelist, value, size);
968 if (flags & CEPH_XATTR_REPLACE)
969 op = CEPH_MDS_OP_RMXATTR;
971 flags |= CEPH_XATTR_REMOVE;
974 dout("setxattr value=%.*s\n", (int)size, value);
977 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
983 req->r_path2 = kstrdup(name, GFP_NOFS);
985 ceph_mdsc_put_request(req);
990 if (op == CEPH_MDS_OP_SETXATTR) {
991 req->r_args.setxattr.flags = cpu_to_le32(flags);
992 req->r_pagelist = pagelist;
996 req->r_inode = inode;
999 req->r_inode_drop = CEPH_CAP_XATTR_SHARED;
1001 dout("xattr.ver (before): %lld\n", ci->i_xattrs.version);
1002 err = ceph_mdsc_do_request(mdsc, NULL, req);
1003 ceph_mdsc_put_request(req);
1004 dout("xattr.ver (after): %lld\n", ci->i_xattrs.version);
1008 ceph_pagelist_release(pagelist);
1012 int __ceph_setxattr(struct inode *inode, const char *name,
1013 const void *value, size_t size, int flags)
1015 struct ceph_vxattr *vxattr;
1016 struct ceph_inode_info *ci = ceph_inode(inode);
1017 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1018 struct ceph_cap_flush *prealloc_cf = NULL;
1019 struct ceph_buffer *old_blob = NULL;
1023 int name_len = strlen(name);
1025 char *newname = NULL;
1026 char *newval = NULL;
1027 struct ceph_inode_xattr *xattr = NULL;
1028 int required_blob_size;
1029 bool check_realm = false;
1030 bool lock_snap_rwsem = false;
1032 if (ceph_snap(inode) != CEPH_NOSNAP)
1035 vxattr = ceph_match_vxattr(inode, name);
1037 if (vxattr->flags & VXATTR_FLAG_READONLY)
1039 if (value && !strncmp(vxattr->name, "ceph.quota", 10))
1043 /* pass any unhandled ceph.* xattrs through to the MDS */
1044 if (!strncmp(name, XATTR_CEPH_PREFIX, XATTR_CEPH_PREFIX_LEN))
1045 goto do_sync_unlocked;
1047 /* preallocate memory for xattr name, value, index node */
1049 newname = kmemdup(name, name_len + 1, GFP_NOFS);
1054 newval = kmemdup(value, val_len, GFP_NOFS);
1059 xattr = kmalloc(sizeof(struct ceph_inode_xattr), GFP_NOFS);
1063 prealloc_cf = ceph_alloc_cap_flush();
1067 spin_lock(&ci->i_ceph_lock);
1069 issued = __ceph_caps_issued(ci, NULL);
1070 if (ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))
1073 if (!lock_snap_rwsem && !ci->i_head_snapc) {
1074 lock_snap_rwsem = true;
1075 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1076 spin_unlock(&ci->i_ceph_lock);
1077 down_read(&mdsc->snap_rwsem);
1078 spin_lock(&ci->i_ceph_lock);
1083 dout("setxattr %p issued %s\n", inode, ceph_cap_string(issued));
1084 __build_xattrs(inode);
1086 required_blob_size = __get_required_blob_size(ci, name_len, val_len);
1088 if (!ci->i_xattrs.prealloc_blob ||
1089 required_blob_size > ci->i_xattrs.prealloc_blob->alloc_len) {
1090 struct ceph_buffer *blob;
1092 spin_unlock(&ci->i_ceph_lock);
1093 ceph_buffer_put(old_blob); /* Shouldn't be required */
1094 dout(" pre-allocating new blob size=%d\n", required_blob_size);
1095 blob = ceph_buffer_new(required_blob_size, GFP_NOFS);
1097 goto do_sync_unlocked;
1098 spin_lock(&ci->i_ceph_lock);
1099 /* prealloc_blob can't be released while holding i_ceph_lock */
1100 if (ci->i_xattrs.prealloc_blob)
1101 old_blob = ci->i_xattrs.prealloc_blob;
1102 ci->i_xattrs.prealloc_blob = blob;
1106 err = __set_xattr(ci, newname, name_len, newval, val_len,
1107 flags, value ? 1 : -1, &xattr);
1110 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_XATTR_EXCL,
1112 ci->i_xattrs.dirty = true;
1113 inode->i_ctime = current_time(inode);
1116 spin_unlock(&ci->i_ceph_lock);
1117 ceph_buffer_put(old_blob);
1118 if (lock_snap_rwsem)
1119 up_read(&mdsc->snap_rwsem);
1121 __mark_inode_dirty(inode, dirty);
1122 ceph_free_cap_flush(prealloc_cf);
1126 spin_unlock(&ci->i_ceph_lock);
1128 if (lock_snap_rwsem)
1129 up_read(&mdsc->snap_rwsem);
1131 /* security module set xattr while filling trace */
1132 if (current->journal_info) {
1133 pr_warn_ratelimited("sync setxattr %p "
1134 "during filling trace\n", inode);
1137 err = ceph_sync_setxattr(inode, name, value, size, flags);
1138 if (err >= 0 && check_realm) {
1139 /* check if snaprealm was created for quota inode */
1140 spin_lock(&ci->i_ceph_lock);
1141 if ((ci->i_max_files || ci->i_max_bytes) &&
1142 !(ci->i_snap_realm &&
1143 ci->i_snap_realm->ino == ci->i_vino.ino))
1145 spin_unlock(&ci->i_ceph_lock);
1149 ceph_free_cap_flush(prealloc_cf);
1156 static int ceph_get_xattr_handler(const struct xattr_handler *handler,
1157 struct dentry *dentry, struct inode *inode,
1158 const char *name, void *value, size_t size)
1160 if (!ceph_is_valid_xattr(name))
1162 return __ceph_getxattr(inode, name, value, size);
1165 static int ceph_set_xattr_handler(const struct xattr_handler *handler,
1166 struct dentry *unused, struct inode *inode,
1167 const char *name, const void *value,
1168 size_t size, int flags)
1170 if (!ceph_is_valid_xattr(name))
1172 return __ceph_setxattr(inode, name, value, size, flags);
1175 static const struct xattr_handler ceph_other_xattr_handler = {
1176 .prefix = "", /* match any name => handlers called with full name */
1177 .get = ceph_get_xattr_handler,
1178 .set = ceph_set_xattr_handler,
1181 #ifdef CONFIG_SECURITY
1182 bool ceph_security_xattr_wanted(struct inode *in)
1184 return in->i_security != NULL;
1187 bool ceph_security_xattr_deadlock(struct inode *in)
1189 struct ceph_inode_info *ci;
1191 if (!in->i_security)
1193 ci = ceph_inode(in);
1194 spin_lock(&ci->i_ceph_lock);
1195 ret = !(ci->i_ceph_flags & CEPH_I_SEC_INITED) &&
1196 !(ci->i_xattrs.version > 0 &&
1197 __ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0));
1198 spin_unlock(&ci->i_ceph_lock);