1 /*
2 * Copyright 2012 Xyratex Technology Limited
3 *
4 * Author: Andrew Perepechko <Andrew_Perepechko@xyratex.com>
5 *
6 */
7
8 #define DEBUG_SUBSYSTEM S_LLITE
9
10 #include <linux/fs.h>
11 #include <linux/sched.h>
12 #include <linux/mm.h>
13 #include "../include/obd_support.h"
14 #include "../include/lustre_lite.h"
15 #include "../include/lustre_dlm.h"
16 #include "../include/lustre_ver.h"
17 #include "llite_internal.h"
18
19 /* If we ever have hundreds of extended attributes, we might want to consider
20 * using a hash or a tree structure instead of list for faster lookups.
21 */
22 struct ll_xattr_entry {
23 struct list_head xe_list; /* protected with
24 * lli_xattrs_list_rwsem */
25 char *xe_name; /* xattr name, \0-terminated */
26 char *xe_value; /* xattr value */
27 unsigned xe_namelen; /* strlen(xe_name) + 1 */
28 unsigned xe_vallen; /* xattr value length */
29 };
30
31 static struct kmem_cache *xattr_kmem;
32 static struct lu_kmem_descr xattr_caches[] = {
33 {
34 .ckd_cache = &xattr_kmem,
35 .ckd_name = "xattr_kmem",
36 .ckd_size = sizeof(struct ll_xattr_entry)
37 },
38 {
39 .ckd_cache = NULL
40 }
41 };
42
ll_xattr_init(void)43 int ll_xattr_init(void)
44 {
45 return lu_kmem_init(xattr_caches);
46 }
47
ll_xattr_fini(void)48 void ll_xattr_fini(void)
49 {
50 lu_kmem_fini(xattr_caches);
51 }
52
53 /**
54 * Initializes xattr cache for an inode.
55 *
56 * This initializes the xattr list and marks cache presence.
57 */
ll_xattr_cache_init(struct ll_inode_info * lli)58 static void ll_xattr_cache_init(struct ll_inode_info *lli)
59 {
60
61 LASSERT(lli != NULL);
62
63 INIT_LIST_HEAD(&lli->lli_xattrs);
64 lli->lli_flags |= LLIF_XATTR_CACHE;
65 }
66
67 /**
68 * This looks for a specific extended attribute.
69 *
70 * Find in @cache and return @xattr_name attribute in @xattr,
71 * for the NULL @xattr_name return the first cached @xattr.
72 *
73 * \retval 0 success
74 * \retval -ENODATA if not found
75 */
ll_xattr_cache_find(struct list_head * cache,const char * xattr_name,struct ll_xattr_entry ** xattr)76 static int ll_xattr_cache_find(struct list_head *cache,
77 const char *xattr_name,
78 struct ll_xattr_entry **xattr)
79 {
80 struct ll_xattr_entry *entry;
81
82 list_for_each_entry(entry, cache, xe_list) {
83 /* xattr_name == NULL means look for any entry */
84 if (xattr_name == NULL ||
85 strcmp(xattr_name, entry->xe_name) == 0) {
86 *xattr = entry;
87 CDEBUG(D_CACHE, "find: [%s]=%.*s\n",
88 entry->xe_name, entry->xe_vallen,
89 entry->xe_value);
90 return 0;
91 }
92 }
93
94 return -ENODATA;
95 }
96
97 /**
98 * This adds an xattr.
99 *
100 * Add @xattr_name attr with @xattr_val value and @xattr_val_len length,
101 *
102 * \retval 0 success
103 * \retval -ENOMEM if no memory could be allocated for the cached attr
104 * \retval -EPROTO if duplicate xattr is being added
105 */
ll_xattr_cache_add(struct list_head * cache,const char * xattr_name,const char * xattr_val,unsigned xattr_val_len)106 static int ll_xattr_cache_add(struct list_head *cache,
107 const char *xattr_name,
108 const char *xattr_val,
109 unsigned xattr_val_len)
110 {
111 struct ll_xattr_entry *xattr;
112
113 if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
114 CDEBUG(D_CACHE, "duplicate xattr: [%s]\n", xattr_name);
115 return -EPROTO;
116 }
117
118 xattr = kmem_cache_alloc(xattr_kmem, GFP_NOFS | __GFP_ZERO);
119 if (xattr == NULL) {
120 CDEBUG(D_CACHE, "failed to allocate xattr\n");
121 return -ENOMEM;
122 }
123
124 xattr->xe_name = kstrdup(xattr_name, GFP_NOFS);
125 if (!xattr->xe_name) {
126 CDEBUG(D_CACHE, "failed to alloc xattr name %u\n",
127 xattr->xe_namelen);
128 goto err_name;
129 }
130 xattr->xe_value = kmemdup(xattr_val, xattr_val_len, GFP_NOFS);
131 if (!xattr->xe_value)
132 goto err_value;
133
134 xattr->xe_vallen = xattr_val_len;
135 list_add(&xattr->xe_list, cache);
136
137 CDEBUG(D_CACHE, "set: [%s]=%.*s\n", xattr_name,
138 xattr_val_len, xattr_val);
139
140 return 0;
141 err_value:
142 kfree(xattr->xe_name);
143 err_name:
144 kmem_cache_free(xattr_kmem, xattr);
145
146 return -ENOMEM;
147 }
148
149 /**
150 * This removes an extended attribute from cache.
151 *
152 * Remove @xattr_name attribute from @cache.
153 *
154 * \retval 0 success
155 * \retval -ENODATA if @xattr_name is not cached
156 */
ll_xattr_cache_del(struct list_head * cache,const char * xattr_name)157 static int ll_xattr_cache_del(struct list_head *cache,
158 const char *xattr_name)
159 {
160 struct ll_xattr_entry *xattr;
161
162 CDEBUG(D_CACHE, "del xattr: %s\n", xattr_name);
163
164 if (ll_xattr_cache_find(cache, xattr_name, &xattr) == 0) {
165 list_del(&xattr->xe_list);
166 kfree(xattr->xe_name);
167 kfree(xattr->xe_value);
168 kmem_cache_free(xattr_kmem, xattr);
169
170 return 0;
171 }
172
173 return -ENODATA;
174 }
175
176 /**
177 * This iterates cached extended attributes.
178 *
179 * Walk over cached attributes in @cache and
180 * fill in @xld_buffer or only calculate buffer
181 * size if @xld_buffer is NULL.
182 *
183 * \retval >= 0 buffer list size
184 * \retval -ENODATA if the list cannot fit @xld_size buffer
185 */
ll_xattr_cache_list(struct list_head * cache,char * xld_buffer,int xld_size)186 static int ll_xattr_cache_list(struct list_head *cache,
187 char *xld_buffer,
188 int xld_size)
189 {
190 struct ll_xattr_entry *xattr, *tmp;
191 int xld_tail = 0;
192
193 list_for_each_entry_safe(xattr, tmp, cache, xe_list) {
194 CDEBUG(D_CACHE, "list: buffer=%p[%d] name=%s\n",
195 xld_buffer, xld_tail, xattr->xe_name);
196
197 if (xld_buffer) {
198 xld_size -= xattr->xe_namelen;
199 if (xld_size < 0)
200 break;
201 memcpy(&xld_buffer[xld_tail],
202 xattr->xe_name, xattr->xe_namelen);
203 }
204 xld_tail += xattr->xe_namelen;
205 }
206
207 if (xld_size < 0)
208 return -ERANGE;
209
210 return xld_tail;
211 }
212
213 /**
214 * Check if the xattr cache is initialized (filled).
215 *
216 * \retval 0 @cache is not initialized
217 * \retval 1 @cache is initialized
218 */
ll_xattr_cache_valid(struct ll_inode_info * lli)219 static int ll_xattr_cache_valid(struct ll_inode_info *lli)
220 {
221 return !!(lli->lli_flags & LLIF_XATTR_CACHE);
222 }
223
224 /**
225 * This finalizes the xattr cache.
226 *
227 * Free all xattr memory. @lli is the inode info pointer.
228 *
229 * \retval 0 no error occurred
230 */
ll_xattr_cache_destroy_locked(struct ll_inode_info * lli)231 static int ll_xattr_cache_destroy_locked(struct ll_inode_info *lli)
232 {
233
234 if (!ll_xattr_cache_valid(lli))
235 return 0;
236
237 while (ll_xattr_cache_del(&lli->lli_xattrs, NULL) == 0)
238 ; /* empty loop */
239 lli->lli_flags &= ~LLIF_XATTR_CACHE;
240
241 return 0;
242 }
243
ll_xattr_cache_destroy(struct inode * inode)244 int ll_xattr_cache_destroy(struct inode *inode)
245 {
246 struct ll_inode_info *lli = ll_i2info(inode);
247 int rc;
248
249 down_write(&lli->lli_xattrs_list_rwsem);
250 rc = ll_xattr_cache_destroy_locked(lli);
251 up_write(&lli->lli_xattrs_list_rwsem);
252
253 return rc;
254 }
255
256 /**
257 * Match or enqueue a PR lock.
258 *
259 * Find or request an LDLM lock with xattr data.
260 * Since LDLM does not provide API for atomic match_or_enqueue,
261 * the function handles it with a separate enq lock.
262 * If successful, the function exits with the list lock held.
263 *
264 * \retval 0 no error occurred
265 * \retval -ENOMEM not enough memory
266 */
ll_xattr_find_get_lock(struct inode * inode,struct lookup_intent * oit,struct ptlrpc_request ** req)267 static int ll_xattr_find_get_lock(struct inode *inode,
268 struct lookup_intent *oit,
269 struct ptlrpc_request **req)
270 {
271 ldlm_mode_t mode;
272 struct lustre_handle lockh = { 0 };
273 struct md_op_data *op_data;
274 struct ll_inode_info *lli = ll_i2info(inode);
275 struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS,
276 .ei_mode = it_to_lock_mode(oit),
277 .ei_cb_bl = ll_md_blocking_ast,
278 .ei_cb_cp = ldlm_completion_ast };
279 struct ll_sb_info *sbi = ll_i2sbi(inode);
280 struct obd_export *exp = sbi->ll_md_exp;
281 int rc;
282
283 mutex_lock(&lli->lli_xattrs_enq_lock);
284 /* inode may have been shrunk and recreated, so data is gone, match lock
285 * only when data exists. */
286 if (ll_xattr_cache_valid(lli)) {
287 /* Try matching first. */
288 mode = ll_take_md_lock(inode, MDS_INODELOCK_XATTR, &lockh, 0,
289 LCK_PR);
290 if (mode != 0) {
291 /* fake oit in mdc_revalidate_lock() manner */
292 oit->d.lustre.it_lock_handle = lockh.cookie;
293 oit->d.lustre.it_lock_mode = mode;
294 goto out;
295 }
296 }
297
298 /* Enqueue if the lock isn't cached locally. */
299 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
300 LUSTRE_OPC_ANY, NULL);
301 if (IS_ERR(op_data)) {
302 mutex_unlock(&lli->lli_xattrs_enq_lock);
303 return PTR_ERR(op_data);
304 }
305
306 op_data->op_valid = OBD_MD_FLXATTR | OBD_MD_FLXATTRLS;
307
308 rc = md_enqueue(exp, &einfo, oit, op_data, &lockh, NULL, 0, NULL, 0);
309 ll_finish_md_op_data(op_data);
310
311 if (rc < 0) {
312 CDEBUG(D_CACHE,
313 "md_intent_lock failed with %d for fid "DFID"\n",
314 rc, PFID(ll_inode2fid(inode)));
315 mutex_unlock(&lli->lli_xattrs_enq_lock);
316 return rc;
317 }
318
319 *req = (struct ptlrpc_request *)oit->d.lustre.it_data;
320 out:
321 down_write(&lli->lli_xattrs_list_rwsem);
322 mutex_unlock(&lli->lli_xattrs_enq_lock);
323
324 return 0;
325 }
326
327 /**
328 * Refill the xattr cache.
329 *
330 * Fetch and cache the whole of xattrs for @inode, acquiring
331 * a read or a write xattr lock depending on operation in @oit.
332 * Intent is dropped on exit unless the operation is setxattr.
333 *
334 * \retval 0 no error occurred
335 * \retval -EPROTO network protocol error
336 * \retval -ENOMEM not enough memory for the cache
337 */
ll_xattr_cache_refill(struct inode * inode,struct lookup_intent * oit)338 static int ll_xattr_cache_refill(struct inode *inode, struct lookup_intent *oit)
339 {
340 struct ll_sb_info *sbi = ll_i2sbi(inode);
341 struct ptlrpc_request *req = NULL;
342 const char *xdata, *xval, *xtail, *xvtail;
343 struct ll_inode_info *lli = ll_i2info(inode);
344 struct mdt_body *body;
345 __u32 *xsizes;
346 int rc, i;
347
348 rc = ll_xattr_find_get_lock(inode, oit, &req);
349 if (rc)
350 goto out_no_unlock;
351
352 /* Do we have the data at this point? */
353 if (ll_xattr_cache_valid(lli)) {
354 ll_stats_ops_tally(sbi, LPROC_LL_GETXATTR_HITS, 1);
355 rc = 0;
356 goto out_maybe_drop;
357 }
358
359 /* Matched but no cache? Cancelled on error by a parallel refill. */
360 if (unlikely(req == NULL)) {
361 CDEBUG(D_CACHE, "cancelled by a parallel getxattr\n");
362 rc = -EIO;
363 goto out_maybe_drop;
364 }
365
366 if (oit->d.lustre.it_status < 0) {
367 CDEBUG(D_CACHE, "getxattr intent returned %d for fid "DFID"\n",
368 oit->d.lustre.it_status, PFID(ll_inode2fid(inode)));
369 rc = oit->d.lustre.it_status;
370 /* xattr data is so large that we don't want to cache it */
371 if (rc == -ERANGE)
372 rc = -EAGAIN;
373 goto out_destroy;
374 }
375
376 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
377 if (body == NULL) {
378 CERROR("no MDT BODY in the refill xattr reply\n");
379 rc = -EPROTO;
380 goto out_destroy;
381 }
382 /* do not need swab xattr data */
383 xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
384 body->eadatasize);
385 xval = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS,
386 body->aclsize);
387 xsizes = req_capsule_server_sized_get(&req->rq_pill, &RMF_EAVALS_LENS,
388 body->max_mdsize * sizeof(__u32));
389 if (xdata == NULL || xval == NULL || xsizes == NULL) {
390 CERROR("wrong setxattr reply\n");
391 rc = -EPROTO;
392 goto out_destroy;
393 }
394
395 xtail = xdata + body->eadatasize;
396 xvtail = xval + body->aclsize;
397
398 CDEBUG(D_CACHE, "caching: xdata=%p xtail=%p\n", xdata, xtail);
399
400 ll_xattr_cache_init(lli);
401
402 for (i = 0; i < body->max_mdsize; i++) {
403 CDEBUG(D_CACHE, "caching [%s]=%.*s\n", xdata, *xsizes, xval);
404 /* Perform consistency checks: attr names and vals in pill */
405 if (memchr(xdata, 0, xtail - xdata) == NULL) {
406 CERROR("xattr protocol violation (names are broken)\n");
407 rc = -EPROTO;
408 } else if (xval + *xsizes > xvtail) {
409 CERROR("xattr protocol violation (vals are broken)\n");
410 rc = -EPROTO;
411 } else if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_XATTR_ENOMEM)) {
412 rc = -ENOMEM;
413 } else if (!strcmp(xdata, XATTR_NAME_ACL_ACCESS)) {
414 /* Filter out ACL ACCESS since it's cached separately */
415 CDEBUG(D_CACHE, "not caching %s\n",
416 XATTR_NAME_ACL_ACCESS);
417 rc = 0;
418 } else {
419 rc = ll_xattr_cache_add(&lli->lli_xattrs, xdata, xval,
420 *xsizes);
421 }
422 if (rc < 0) {
423 ll_xattr_cache_destroy_locked(lli);
424 goto out_destroy;
425 }
426 xdata += strlen(xdata) + 1;
427 xval += *xsizes;
428 xsizes++;
429 }
430
431 if (xdata != xtail || xval != xvtail)
432 CERROR("a hole in xattr data\n");
433
434 ll_set_lock_data(sbi->ll_md_exp, inode, oit, NULL);
435
436 goto out_maybe_drop;
437 out_maybe_drop:
438
439 ll_intent_drop_lock(oit);
440
441 if (rc != 0)
442 up_write(&lli->lli_xattrs_list_rwsem);
443 out_no_unlock:
444 ptlrpc_req_finished(req);
445
446 return rc;
447
448 out_destroy:
449 up_write(&lli->lli_xattrs_list_rwsem);
450
451 ldlm_lock_decref_and_cancel((struct lustre_handle *)
452 &oit->d.lustre.it_lock_handle,
453 oit->d.lustre.it_lock_mode);
454
455 goto out_no_unlock;
456 }
457
458 /**
459 * Get an xattr value or list xattrs using the write-through cache.
460 *
461 * Get the xattr value (@valid has OBD_MD_FLXATTR set) of @name or
462 * list xattr names (@valid has OBD_MD_FLXATTRLS set) for @inode.
463 * The resulting value/list is stored in @buffer if the former
464 * is not larger than @size.
465 *
466 * \retval 0 no error occurred
467 * \retval -EPROTO network protocol error
468 * \retval -ENOMEM not enough memory for the cache
469 * \retval -ERANGE the buffer is not large enough
470 * \retval -ENODATA no such attr or the list is empty
471 */
ll_xattr_cache_get(struct inode * inode,const char * name,char * buffer,size_t size,__u64 valid)472 int ll_xattr_cache_get(struct inode *inode,
473 const char *name,
474 char *buffer,
475 size_t size,
476 __u64 valid)
477 {
478 struct lookup_intent oit = { .it_op = IT_GETXATTR };
479 struct ll_inode_info *lli = ll_i2info(inode);
480 int rc = 0;
481
482 LASSERT(!!(valid & OBD_MD_FLXATTR) ^ !!(valid & OBD_MD_FLXATTRLS));
483
484 down_read(&lli->lli_xattrs_list_rwsem);
485 if (!ll_xattr_cache_valid(lli)) {
486 up_read(&lli->lli_xattrs_list_rwsem);
487 rc = ll_xattr_cache_refill(inode, &oit);
488 if (rc)
489 return rc;
490 downgrade_write(&lli->lli_xattrs_list_rwsem);
491 } else {
492 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR_HITS, 1);
493 }
494
495 if (valid & OBD_MD_FLXATTR) {
496 struct ll_xattr_entry *xattr;
497
498 rc = ll_xattr_cache_find(&lli->lli_xattrs, name, &xattr);
499 if (rc == 0) {
500 rc = xattr->xe_vallen;
501 /* zero size means we are only requested size in rc */
502 if (size != 0) {
503 if (size >= xattr->xe_vallen)
504 memcpy(buffer, xattr->xe_value,
505 xattr->xe_vallen);
506 else
507 rc = -ERANGE;
508 }
509 }
510 } else if (valid & OBD_MD_FLXATTRLS) {
511 rc = ll_xattr_cache_list(&lli->lli_xattrs,
512 size ? buffer : NULL, size);
513 }
514
515 goto out;
516 out:
517 up_read(&lli->lli_xattrs_list_rwsem);
518
519 return rc;
520 }
521