3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 only,
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License version 2 for more details (a copy is included
13 * in the LICENSE file that accompanied this code).
15 * You should have received a copy of the GNU General Public License
16 * version 2 along with this program; If not, see http://www.gnu.org/licenses
18 * Please visit http://www.xyratex.com/contact if you need additional
19 * information or have any questions.
25 * Copyright 2012 Xyratex Technology Limited
27 * Copyright (c) 2013, 2017, Intel Corporation.
29 * Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
33 #define DEBUG_SUBSYSTEM S_LLITE
36 #include <linux/sched.h>
38 #include <obd_support.h>
39 #include <lustre_dlm.h>
40 #include "llite_internal.h"
42 /* If we ever have hundreds of extended attributes, we might want to consider
43 * using a hash or a tree structure instead of list for faster lookups.
45 struct ll_xattr_entry {
46 struct list_head xe_list; /* protected with
47 * lli_xattrs_list_rwsem */
48 char *xe_name; /* xattr name, \0-terminated */
49 char *xe_value; /* xattr value */
50 unsigned xe_namelen; /* strlen(xe_name) + 1 */
51 unsigned xe_vallen; /* xattr value length */
54 static struct kmem_cache *xattr_kmem;
55 static struct lu_kmem_descr xattr_caches[] = {
57 .ckd_cache = &xattr_kmem,
58 .ckd_name = "xattr_kmem",
59 .ckd_size = sizeof(struct ll_xattr_entry)
66 int ll_xattr_init(void)
68 return lu_kmem_init(xattr_caches);
71 void ll_xattr_fini(void)
73 lu_kmem_fini(xattr_caches);
77 * Initializes xattr cache for an inode.
79 * This initializes the xattr list and marks cache presence.
81 static void ll_xattr_cache_init(struct ll_inode_info *lli)
87 INIT_LIST_HEAD(&lli->lli_xattrs);
88 set_bit(LLIF_XATTR_CACHE, &lli->lli_flags);
92 * This looks for a specific extended attribute.
94 * Find in @cache and return @xattr_name attribute in @xattr,
95 * for the NULL @xattr_name return the first cached @xattr.
98 * \retval -ENODATA if not found
100 static int ll_xattr_cache_find(struct list_head *cache,
101 const char *xattr_name,
102 struct ll_xattr_entry **xattr)
104 struct ll_xattr_entry *entry;
108 list_for_each_entry(entry, cache, xe_list) {
109 /* xattr_name == NULL means look for any entry */
110 if (xattr_name == NULL ||
111 strcmp(xattr_name, entry->xe_name) == 0) {
113 CDEBUG(D_CACHE, "find: [%s]=%.*s\n",
114 entry->xe_name, entry->xe_vallen,
124 * This adds an xattr.
126 * Add @xattr_name attr with @xattr_val value and @xattr_val_len length,
129 * \retval -ENOMEM if no memory could be allocated for the cached attr
130 * \retval -EPROTO if duplicate xattr is being added
132 static int ll_xattr_cache_add(struct list_head *cache,
133 const char *xattr_name,
134 const char *xattr_val,
135 unsigned xattr_val_len)
137 struct ll_xattr_entry *xattr;
141 if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
142 if (!strcmp(xattr_name, LL_XATTR_NAME_ENCRYPTION_CONTEXT) ||
143 !strcmp(xattr_name, LL_XATTR_NAME_ENCRYPTION_CONTEXT_OLD))
144 /* it means enc ctx was already in cache,
145 * ignore error as it cannot be modified
149 CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
153 OBD_SLAB_ALLOC_PTR_GFP(xattr, xattr_kmem, GFP_NOFS);
155 CDEBUG(D_CACHE, "failed to allocate xattr\n");
159 xattr->xe_namelen = strlen(xattr_name) + 1;
161 OBD_ALLOC(xattr->xe_name, xattr->xe_namelen);
162 if (!xattr->xe_name) {
163 CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
167 OBD_ALLOC(xattr->xe_value, xattr_val_len);
168 if (!xattr->xe_value) {
169 CDEBUG(D_CACHE, "failed to alloc xattr value %d\n",
174 memcpy(xattr->xe_name, xattr_name, xattr->xe_namelen);
175 memcpy(xattr->xe_value, xattr_val, xattr_val_len);
176 xattr->xe_vallen = xattr_val_len;
177 list_add(&xattr->xe_list, cache);
179 CDEBUG(D_CACHE, "set: [%s]=%.*s\n", xattr_name,
180 xattr_val_len, xattr_val);
184 OBD_FREE(xattr->xe_name, xattr->xe_namelen);
186 OBD_SLAB_FREE_PTR(xattr, xattr_kmem);
192 * This removes an extended attribute from cache.
194 * Remove @xattr_name attribute from @cache.
197 * \retval -ENODATA if @xattr_name is not cached
199 static int ll_xattr_cache_del(struct list_head *cache,
200 const char *xattr_name)
202 struct ll_xattr_entry *xattr;
206 CDEBUG(D_CACHE, "del xattr: %s\n", xattr_name);
208 if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
209 list_del(&xattr->xe_list);
210 OBD_FREE(xattr->xe_name, xattr->xe_namelen);
211 OBD_FREE(xattr->xe_value, xattr->xe_vallen);
212 OBD_SLAB_FREE_PTR(xattr, xattr_kmem);
221 * This iterates cached extended attributes.
223 * Walk over cached attributes in @cache and
224 * fill in @xld_buffer or only calculate buffer
225 * size if @xld_buffer is NULL.
227 * \retval >= 0 buffer list size
228 * \retval -ENODATA if the list cannot fit @xld_size buffer
230 static int ll_xattr_cache_list(struct list_head *cache,
234 struct ll_xattr_entry *xattr, *tmp;
239 list_for_each_entry_safe(xattr, tmp, cache, xe_list) {
240 CDEBUG(D_CACHE, "list: buffer=%p[%d] name=%s\n",
241 xld_buffer, xld_tail, xattr->xe_name);
244 xld_size -= xattr->xe_namelen;
247 memcpy(&xld_buffer[xld_tail],
248 xattr->xe_name, xattr->xe_namelen);
250 xld_tail += xattr->xe_namelen;
260 * Check if the xattr cache is initialized.
262 * \retval 0 @cache is not initialized
263 * \retval 1 @cache is initialized
265 static int ll_xattr_cache_valid(struct ll_inode_info *lli)
267 return test_bit(LLIF_XATTR_CACHE, &lli->lli_flags);
271 * Check if the xattr cache is filled.
273 * \retval 0 @cache is not filled
274 * \retval 1 @cache is filled
276 static int ll_xattr_cache_filled(struct ll_inode_info *lli)
278 return test_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
282 * This finalizes the xattr cache.
284 * Free all xattr memory. @lli is the inode info pointer.
286 * \retval 0 no error occured
288 static int ll_xattr_cache_destroy_locked(struct ll_inode_info *lli)
292 if (!ll_xattr_cache_valid(lli))
295 while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
298 clear_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
299 clear_bit(LLIF_XATTR_CACHE, &lli->lli_flags);
304 int ll_xattr_cache_destroy(struct inode *inode)
306 struct ll_inode_info *lli = ll_i2info(inode);
311 down_write(&lli->lli_xattrs_list_rwsem);
312 rc = ll_xattr_cache_destroy_locked(lli);
313 up_write(&lli->lli_xattrs_list_rwsem);
319 * ll_xattr_cache_empty - empty xattr cache for @ino
321 * Similar to ll_xattr_cache_destroy(), but preserves encryption context.
322 * So only LLIF_XATTR_CACHE_FILLED flag is cleared, but not LLIF_XATTR_CACHE.
324 int ll_xattr_cache_empty(struct inode *inode)
326 struct ll_inode_info *lli = ll_i2info(inode);
327 struct ll_xattr_entry *entry, *n;
331 down_write(&lli->lli_xattrs_list_rwsem);
332 if (!ll_xattr_cache_valid(lli) ||
333 !ll_xattr_cache_filled(lli))
336 list_for_each_entry_safe(entry, n, &lli->lli_xattrs, xe_list) {
337 if (strcmp(entry->xe_name, xattr_for_enc(inode)) == 0)
340 CDEBUG(D_CACHE, "delete: %s\n", entry->xe_name);
341 list_del(&entry->xe_list);
342 OBD_FREE(entry->xe_name, entry->xe_namelen);
343 OBD_FREE(entry->xe_value, entry->xe_vallen);
344 OBD_SLAB_FREE_PTR(entry, xattr_kmem);
346 clear_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
349 up_write(&lli->lli_xattrs_list_rwsem);
354 * Match or enqueue a PR lock.
356 * Find or request an LDLM lock with xattr data.
357 * Since LDLM does not provide API for atomic match_or_enqueue,
358 * the function handles it with a separate enq lock.
359 * If successful, the function exits with a write lock held
360 * on lli_xattrs_list_rwsem.
362 * \retval 0 no error occured
363 * \retval -ENOMEM not enough memory
365 static int ll_xattr_find_get_lock(struct inode *inode,
366 struct lookup_intent *oit,
367 struct ptlrpc_request **req)
370 struct lustre_handle lockh = { 0 };
371 struct md_op_data *op_data;
372 struct ll_inode_info *lli = ll_i2info(inode);
373 struct ll_sb_info *sbi = ll_i2sbi(inode);
374 struct obd_export *exp = sbi->ll_md_exp;
379 mutex_lock(&lli->lli_xattrs_enq_lock);
380 /* inode may have been shrunk and recreated, so data is gone, match lock
381 * only when data exists. */
382 if (ll_xattr_cache_filled(lli)) {
383 /* Try matching first. */
384 mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0,
387 /* fake oit in mdc_revalidate_lock() manner */
388 oit->it_lock_handle = lockh.cookie;
389 oit->it_lock_mode = mode;
394 /* Enqueue if the lock isn't cached locally. */
395 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
396 LUSTRE_OPC_ANY, NULL);
397 if (IS_ERR(op_data)) {
398 mutex_unlock(&lli->lli_xattrs_enq_lock);
399 RETURN(PTR_ERR(op_data));
402 op_data->op_valid = OBD_MD_FLXATTR | OBD_MD_FLXATTRLS;
404 rc = md_intent_lock(exp, op_data, oit, req, &ll_md_blocking_ast, 0);
405 ll_finish_md_op_data(op_data);
406 *req = oit->it_request;
409 CDEBUG(D_CACHE, "md_intent_lock failed with %d for fid "DFID"\n",
410 rc, PFID(ll_inode2fid(inode)));
411 mutex_unlock(&lli->lli_xattrs_enq_lock);
416 down_write(&lli->lli_xattrs_list_rwsem);
417 mutex_unlock(&lli->lli_xattrs_enq_lock);
423 * Refill the xattr cache.
425 * Fetch and cache the whole of xattrs for @inode, thanks to the write lock
426 * on lli_xattrs_list_rwsem obtained from ll_xattr_find_get_lock().
427 * If successful, this write lock is kept.
429 * \retval 0 no error occured
430 * \retval -EPROTO network protocol error
431 * \retval -ENOMEM not enough memory for the cache
433 static int ll_xattr_cache_refill(struct inode *inode)
435 struct lookup_intent oit = { .it_op = IT_GETXATTR };
436 struct ll_sb_info *sbi = ll_i2sbi(inode);
437 struct ptlrpc_request *req = NULL;
438 const char *xdata, *xval, *xtail, *xvtail;
439 struct ll_inode_info *lli = ll_i2info(inode);
440 struct mdt_body *body;
446 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_XATTR_PAUSE, cfs_fail_val ?: 2);
448 rc = ll_xattr_find_get_lock(inode, &oit, &req);
452 /* Do we have the data at this point? */
453 if (ll_xattr_cache_filled(lli)) {
454 ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
455 ll_intent_drop_lock(&oit);
456 GOTO(err_req, rc = 0);
459 /* Matched but no cache? Cancelled on error by a parallel refill. */
460 if (unlikely(req == NULL)) {
461 CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
462 ll_intent_drop_lock(&oit);
463 GOTO(err_unlock, rc = -EAGAIN);
466 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
468 CERROR("no MDT BODY in the refill xattr reply\n");
469 GOTO(err_cancel, rc = -EPROTO);
471 /* do not need swab xattr data */
472 xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
473 body->mbo_eadatasize);
474 xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
476 xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
477 body->mbo_max_mdsize *
479 if (xdata == NULL || xval == NULL || xsizes == NULL) {
480 CERROR("wrong setxattr reply\n");
481 GOTO(err_cancel, rc = -EPROTO);
484 xtail = xdata + body->mbo_eadatasize;
485 xvtail = xval + body->mbo_aclsize;
487 CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);
489 if (!ll_xattr_cache_valid(lli))
490 ll_xattr_cache_init(lli);
492 for (i = 0; i < body->mbo_max_mdsize; i++) {
493 CDEBUG(D_CACHE, "caching [%s]=%.*s\n", xdata, *xsizes, xval);
494 /* Perform consistency checks: attr names and vals in pill */
495 if (memchr(xdata, 0, xtail - xdata) == NULL) {
496 CERROR("xattr protocol violation (names are broken)\n");
498 } else if (xval + *xsizes > xvtail) {
499 CERROR("xattr protocol violation (vals are broken)\n");
501 } else if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_XATTR_ENOMEM)) {
503 } else if (!strcmp(xdata, XATTR_NAME_ACL_ACCESS)) {
504 /* Filter out ACL ACCESS since it's cached separately */
505 CDEBUG(D_CACHE, "not caching %s\n",
506 XATTR_NAME_ACL_ACCESS);
508 } else if (ll_xattr_is_seclabel(xdata)) {
509 /* Filter out security label, it is cached in slab */
510 CDEBUG(D_CACHE, "not caching %s\n", xdata);
512 } else if (!strcmp(xdata, XATTR_NAME_SOM)) {
513 /* Filter out trusted.som, it is not cached on client */
514 CDEBUG(D_CACHE, "not caching trusted.som\n");
517 rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
521 ll_xattr_cache_destroy_locked(lli);
522 GOTO(err_cancel, rc);
524 xdata += strlen(xdata) + 1;
529 if (xdata != xtail || xval != xvtail)
530 CERROR("a hole in xattr data\n");
532 set_bit(LLIF_XATTR_CACHE_FILLED, &lli->lli_flags);
534 ll_set_lock_data(sbi->ll_md_exp, inode, &oit, NULL);
535 ll_intent_drop_lock(&oit);
537 ptlrpc_req_finished(req);
541 ldlm_lock_decref_and_cancel((struct lustre_handle *)
545 up_write(&lli->lli_xattrs_list_rwsem);
550 ptlrpc_req_finished(req);
555 * Get an xattr value or list xattrs using the write-through cache.
557 * Get the xattr value (@valid has OBD_MD_FLXATTR set) of @name or
558 * list xattr names (@valid has OBD_MD_FLXATTRLS set) for @inode.
559 * The resulting value/list is stored in @buffer if the former
560 * is not larger than @size.
562 * \retval 0 no error occured
563 * \retval -EPROTO network protocol error
564 * \retval -ENOMEM not enough memory for the cache
565 * \retval -ERANGE the buffer is not large enough
566 * \retval -ENODATA no such attr or the list is empty
568 int ll_xattr_cache_get(struct inode *inode,
574 struct ll_inode_info *lli = ll_i2info(inode);
579 LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));
581 down_read(&lli->lli_xattrs_list_rwsem);
582 /* For performance reasons, we do not want to refill complete xattr
583 * cache if we are just interested in encryption context.
585 if ((valid & OBD_MD_FLXATTRLS ||
586 strcmp(name, xattr_for_enc(inode)) != 0) &&
587 !ll_xattr_cache_filled(lli)) {
588 up_read(&lli->lli_xattrs_list_rwsem);
589 rc = ll_xattr_cache_refill(inode);
592 /* Turn the write lock obtained in ll_xattr_cache_refill()
595 downgrade_write(&lli->lli_xattrs_list_rwsem);
597 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
600 if (!ll_xattr_cache_valid(lli))
601 GOTO(out, rc = -ENODATA);
603 if (valid & OBD_MD_FLXATTR) {
604 struct ll_xattr_entry *xattr;
606 rc = ll_xattr_cache_find(&lli->lli_xattrs, name, &xattr);
608 rc = xattr->xe_vallen;
609 /* zero size means we are only requested size in rc */
611 if (size >= xattr->xe_vallen)
612 memcpy(buffer, xattr->xe_value,
617 /* Return the project id when the virtual project id xattr
618 * is explicitly asked.
620 } else if (strcmp(name, XATTR_NAME_PROJID) == 0) {
621 /* 10 chars to hold u32 in decimal, plus ending \0 */
624 rc = snprintf(projid, sizeof(projid),
625 "%u", lli->lli_projid);
628 memcpy(buffer, projid, rc);
633 } else if (valid & OBD_MD_FLXATTRLS) {
634 rc = ll_xattr_cache_list(&lli->lli_xattrs,
635 size ? buffer : NULL, size);
640 up_read(&lli->lli_xattrs_list_rwsem);
646 * Insert an xattr value into the cache.
648 * Add @name xattr with @buffer value and @size length for @inode.
649 * Init cache for @inode if necessary.
652 * \retval < 0 from ll_xattr_cache_add(), except -EPROTO is ignored for
653 * LL_XATTR_NAME_ENCRYPTION_CONTEXT xattr
655 int ll_xattr_cache_insert(struct inode *inode,
660 struct ll_inode_info *lli = ll_i2info(inode);
665 down_write(&lli->lli_xattrs_list_rwsem);
666 if (!ll_xattr_cache_valid(lli))
667 ll_xattr_cache_init(lli);
668 rc = ll_xattr_cache_add(&lli->lli_xattrs, name, buffer, size);
669 up_write(&lli->lli_xattrs_list_rwsem);