4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
104 ll_generate_random_uuid(uuid);
105 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
106 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
108 sbi->ll_flags |= LL_SBI_VERBOSE;
109 #ifdef ENABLE_CHECKSUM
110 sbi->ll_flags |= LL_SBI_CHECKSUM;
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 /* metadata statahead is enabled by default */
126 sbi->ll_sa_max = LL_SA_RPC_DEF;
127 atomic_set(&sbi->ll_sa_total, 0);
128 atomic_set(&sbi->ll_sa_wrong, 0);
129 atomic_set(&sbi->ll_sa_running, 0);
130 atomic_set(&sbi->ll_agl_total, 0);
131 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
132 sbi->ll_flags |= LL_SBI_FAST_READ;
135 sbi->ll_squash.rsi_uid = 0;
136 sbi->ll_squash.rsi_gid = 0;
137 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
138 init_rwsem(&sbi->ll_squash.rsi_sem);
143 static void ll_free_sbi(struct super_block *sb)
145 struct ll_sb_info *sbi = ll_s2sbi(sb);
149 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
150 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
151 if (sbi->ll_cache != NULL) {
152 cl_cache_decref(sbi->ll_cache);
153 sbi->ll_cache = NULL;
155 OBD_FREE(sbi, sizeof(*sbi));
160 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
162 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
163 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
167 struct vfsmount *mnt)
169 struct inode *root = NULL;
170 struct ll_sb_info *sbi = ll_s2sbi(sb);
171 struct obd_device *obd;
172 struct obd_statfs *osfs = NULL;
173 struct ptlrpc_request *request = NULL;
174 struct obd_connect_data *data = NULL;
175 struct obd_uuid *uuid;
176 struct md_op_data *op_data;
177 struct lustre_md lmd;
179 int size, err, checksum;
182 obd = class_name2obd(md);
184 CERROR("MD %s: not setup or attached\n", md);
198 /* pass client page size via ocd_grant_blkbits, the server should report
199 * back its backend blocksize for grant calculation purpose */
200 data->ocd_grant_blkbits = PAGE_SHIFT;
202 /* indicate MDT features supported by this client */
203 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
204 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
205 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
206 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
207 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
208 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
209 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
210 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
211 OBD_CONNECT_64BITHASH |
212 OBD_CONNECT_EINPROGRESS |
213 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
214 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
215 OBD_CONNECT_MAX_EASIZE |
216 OBD_CONNECT_FLOCK_DEAD |
217 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
218 OBD_CONNECT_OPEN_BY_FID |
219 OBD_CONNECT_DIR_STRIPE |
220 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
221 OBD_CONNECT_SUBTREE |
222 OBD_CONNECT_MULTIMODRPCS |
223 OBD_CONNECT_GRANT_PARAM |
224 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
226 data->ocd_connect_flags2 = OBD_CONNECT2_FLR;
228 #ifdef HAVE_LRU_RESIZE_SUPPORT
229 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
230 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
232 #ifdef CONFIG_FS_POSIX_ACL
233 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
234 OBD_CONNECT_LARGE_ACL;
237 data->ocd_cksum_types = cksum_types_supported_client();
239 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
240 /* flag mdc connection as lightweight, only used for test
241 * purpose, use with care */
242 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
244 data->ocd_ibits_known = MDS_INODELOCK_FULL;
245 data->ocd_version = LUSTRE_VERSION_CODE;
247 if (sb->s_flags & MS_RDONLY)
248 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
249 if (sbi->ll_flags & LL_SBI_USER_XATTR)
250 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
253 /* Setting this indicates we correctly support S_NOSEC (See kernel
254 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
256 sb->s_flags |= MS_NOSEC;
259 if (sbi->ll_flags & LL_SBI_FLOCK)
260 sbi->ll_fop = &ll_file_operations_flock;
261 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
262 sbi->ll_fop = &ll_file_operations;
264 sbi->ll_fop = &ll_file_operations_noflock;
266 /* always ping even if server suppress_pings */
267 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
268 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
270 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
271 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
272 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
274 data->ocd_brw_size = MD_MAX_BRW_SIZE;
276 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
278 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
279 "recovery, of which this client is not a "
280 "part. Please wait for recovery to complete,"
281 " abort, or time out.\n", md);
284 CERROR("cannot connect to %s: rc = %d\n", md, err);
288 sbi->ll_md_exp->exp_connect_data = *data;
290 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
291 LUSTRE_SEQ_METADATA);
293 CERROR("%s: Can't init metadata layer FID infrastructure, "
294 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
298 /* For mount, we only need fs info from MDT0, and also in DNE, it
299 * can make sure the client can be mounted as long as MDT0 is
301 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
302 ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
303 OBD_STATFS_FOR_MDT0);
305 GOTO(out_md_fid, err);
307 /* This needs to be after statfs to ensure connect has finished.
308 * Note that "data" does NOT contain the valid connect reply.
309 * If connecting to a 1.8 server there will be no LMV device, so
310 * we can access the MDC export directly and exp_connect_flags will
311 * be non-zero, but if accessing an upgraded 2.1 server it will
312 * have the correct flags filled in.
313 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
314 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
315 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
316 valid != CLIENT_CONNECT_MDT_REQD) {
319 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
320 obd_connect_flags2str(buf, PAGE_SIZE,
321 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
322 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
323 "feature(s) needed for correct operation "
324 "of this client (%s). Please upgrade "
325 "server or downgrade client.\n",
326 sbi->ll_md_exp->exp_obd->obd_name, buf);
327 OBD_FREE(buf, PAGE_SIZE);
328 GOTO(out_md_fid, err = -EPROTO);
331 size = sizeof(*data);
332 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
333 KEY_CONN_DATA, &size, data);
335 CERROR("%s: Get connect data failed: rc = %d\n",
336 sbi->ll_md_exp->exp_obd->obd_name, err);
337 GOTO(out_md_fid, err);
340 LASSERT(osfs->os_bsize);
341 sb->s_blocksize = osfs->os_bsize;
342 sb->s_blocksize_bits = log2(osfs->os_bsize);
343 sb->s_magic = LL_SUPER_MAGIC;
344 sb->s_maxbytes = MAX_LFS_FILESIZE;
345 sbi->ll_namelen = osfs->os_namelen;
346 sbi->ll_mnt.mnt = current->fs->root.mnt;
348 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
349 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
350 LCONSOLE_INFO("Disabling user_xattr feature because "
351 "it is not supported on the server\n");
352 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
355 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
357 sb->s_flags |= MS_POSIXACL;
359 sbi->ll_flags |= LL_SBI_ACL;
361 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
363 sb->s_flags &= ~MS_POSIXACL;
365 sbi->ll_flags &= ~LL_SBI_ACL;
368 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
369 sbi->ll_flags |= LL_SBI_64BIT_HASH;
371 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
372 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
374 if (obd_connect_has_secctx(data))
375 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
377 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
378 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
379 LCONSOLE_INFO("%s: disabling xattr cache due to "
380 "unknown maximum xattr size.\n", dt);
381 } else if (!sbi->ll_xattr_cache_set) {
382 /* If xattr_cache is already set (no matter 0 or 1)
383 * during processing llog, it won't be enabled here. */
384 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
385 sbi->ll_xattr_cache_enabled = 1;
389 obd = class_name2obd(dt);
391 CERROR("DT %s: not setup or attached\n", dt);
392 GOTO(out_md_fid, err = -ENODEV);
395 /* pass client page size via ocd_grant_blkbits, the server should report
396 * back its backend blocksize for grant calculation purpose */
397 data->ocd_grant_blkbits = PAGE_SHIFT;
399 /* indicate OST features supported by this client */
400 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
401 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
402 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
403 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
404 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
405 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
406 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
407 OBD_CONNECT_EINPROGRESS |
408 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
409 OBD_CONNECT_LAYOUTLOCK |
410 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
411 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
414 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
415 * can interoperate with an older version of lockahead which was released prior
416 * to landing in master. This support will be dropped when 2.13 development
417 * starts. At the point, we should not just drop the connect flag (below), we
418 * should also remove the support in the code.
420 * Removing it means a few things:
421 * 1. Remove this section here
422 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
423 * 3. Remove function exp_connect_lockahead_old
424 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
426 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
427 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
430 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
432 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
433 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
435 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
436 * disabled by default, because it can still be enabled on the
437 * fly via /proc. As a consequence, we still need to come to an
438 * agreement on the supported algorithms at connect time */
439 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
441 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
442 data->ocd_cksum_types = OBD_CKSUM_ADLER;
444 data->ocd_cksum_types = cksum_types_supported_client();
446 #ifdef HAVE_LRU_RESIZE_SUPPORT
447 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
449 /* always ping even if server suppress_pings */
450 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
451 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
453 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
454 "ocd_grant: %d\n", data->ocd_connect_flags,
455 data->ocd_version, data->ocd_grant);
457 obd->obd_upcall.onu_owner = &sbi->ll_lco;
458 obd->obd_upcall.onu_upcall = cl_ocd_update;
460 data->ocd_brw_size = DT_MAX_BRW_SIZE;
462 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
465 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
466 "recovery, of which this client is not a "
467 "part. Please wait for recovery to "
468 "complete, abort, or time out.\n", dt);
471 CERROR("%s: Cannot connect to %s: rc = %d\n",
472 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
476 sbi->ll_dt_exp->exp_connect_data = *data;
478 /* Don't change value if it was specified in the config log */
479 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
480 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
481 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
482 (data->ocd_brw_size >> PAGE_SHIFT));
484 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
485 LUSTRE_SEQ_METADATA);
487 CERROR("%s: Can't init data layer FID infrastructure, "
488 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
492 mutex_lock(&sbi->ll_lco.lco_lock);
493 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
494 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
495 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
496 mutex_unlock(&sbi->ll_lco.lco_lock);
498 fid_zero(&sbi->ll_root_fid);
499 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
502 CERROR("cannot mds_connect: rc = %d\n", err);
503 GOTO(out_lock_cn_cb, err);
505 if (!fid_is_sane(&sbi->ll_root_fid)) {
506 CERROR("%s: Invalid root fid "DFID" during mount\n",
507 sbi->ll_md_exp->exp_obd->obd_name,
508 PFID(&sbi->ll_root_fid));
509 GOTO(out_lock_cn_cb, err = -EINVAL);
511 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
513 sb->s_op = &lustre_super_operations;
514 #ifdef HAVE_XATTR_HANDLER_FLAGS
515 sb->s_xattr = ll_xattr_handlers;
517 #if THREAD_SIZE >= 8192 /*b=17630*/
518 sb->s_export_op = &lustre_export_operations;
522 * XXX: move this to after cbd setup? */
523 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
524 if (sbi->ll_flags & LL_SBI_ACL)
525 valid |= OBD_MD_FLACL;
527 OBD_ALLOC_PTR(op_data);
529 GOTO(out_lock_cn_cb, err = -ENOMEM);
531 op_data->op_fid1 = sbi->ll_root_fid;
532 op_data->op_mode = 0;
533 op_data->op_valid = valid;
535 err = md_getattr(sbi->ll_md_exp, op_data, &request);
537 OBD_FREE_PTR(op_data);
539 CERROR("%s: md_getattr failed for root: rc = %d\n",
540 sbi->ll_md_exp->exp_obd->obd_name, err);
541 GOTO(out_lock_cn_cb, err);
544 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
545 sbi->ll_md_exp, &lmd);
547 CERROR("failed to understand root inode md: rc = %d\n", err);
548 ptlrpc_req_finished(request);
549 GOTO(out_lock_cn_cb, err);
552 LASSERT(fid_is_sane(&sbi->ll_root_fid));
553 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
554 sbi->ll_flags & LL_SBI_32BIT_API),
556 md_free_lustre_md(sbi->ll_md_exp, &lmd);
557 ptlrpc_req_finished(request);
560 #ifdef CONFIG_FS_POSIX_ACL
562 posix_acl_release(lmd.posix_acl);
563 lmd.posix_acl = NULL;
566 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
568 CERROR("lustre_lite: bad iget4 for root\n");
572 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
573 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
574 KEY_CHECKSUM, sizeof(checksum), &checksum,
577 CERROR("%s: Set checksum failed: rc = %d\n",
578 sbi->ll_dt_exp->exp_obd->obd_name, err);
583 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
584 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
585 sbi->ll_cache, NULL);
587 CERROR("%s: Set cache_set failed: rc = %d\n",
588 sbi->ll_dt_exp->exp_obd->obd_name, err);
592 sb->s_root = d_make_root(root);
593 if (sb->s_root == NULL) {
594 CERROR("%s: can't make root dentry\n",
595 ll_get_fsname(sb, NULL, 0));
596 GOTO(out_root, err = -ENOMEM);
598 #ifdef HAVE_DCACHE_LOCK
599 sb->s_root->d_op = &ll_d_ops;
602 sbi->ll_sdev_orig = sb->s_dev;
604 /* We set sb->s_dev equal on all lustre clients in order to support
605 * NFS export clustering. NFSD requires that the FSID be the same
607 /* s_dev is also used in lt_compare() to compare two fs, but that is
608 * only a node-local comparison. */
609 uuid = obd_get_uuid(sbi->ll_md_exp);
611 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
618 if (sbi->ll_proc_root != NULL) {
619 err = lprocfs_ll_register_obd(sb, dt);
621 CERROR("%s: could not register %s in llite: rc = %d\n",
622 dt, ll_get_fsname(sb, NULL, 0), err);
625 err = lprocfs_ll_register_obd(sb, md);
627 CERROR("%s: could not register %s in llite: rc = %d\n",
628 md, ll_get_fsname(sb, NULL, 0), err);
638 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
640 obd_disconnect(sbi->ll_dt_exp);
641 sbi->ll_dt_exp = NULL;
643 obd_fid_fini(sbi->ll_md_exp->exp_obd);
645 obd_disconnect(sbi->ll_md_exp);
646 sbi->ll_md_exp = NULL;
655 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
659 size = sizeof(*lmmsize);
660 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
661 KEY_MAX_EASIZE, &size, lmmsize);
663 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
664 sbi->ll_dt_exp->exp_obd->obd_name, rc);
669 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
670 KEY_MAX_EASIZE, &size, lmmsize);
672 CERROR("Get max mdsize error rc %d\n", rc);
678 * Get the value of the default_easize parameter.
680 * \see client_obd::cl_default_mds_easize
682 * \param[in] sbi superblock info for this filesystem
683 * \param[out] lmmsize pointer to storage location for value
685 * \retval 0 on success
686 * \retval negative negated errno on failure
688 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
693 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
694 KEY_DEFAULT_EASIZE, &size, lmmsize);
696 CERROR("Get default mdsize error rc %d\n", rc);
702 * Set the default_easize parameter to the given value.
704 * \see client_obd::cl_default_mds_easize
706 * \param[in] sbi superblock info for this filesystem
707 * \param[in] lmmsize the size to set
709 * \retval 0 on success
710 * \retval negative negated errno on failure
712 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
716 if (lmmsize < sizeof(struct lov_mds_md) ||
717 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
720 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
721 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
722 sizeof(int), &lmmsize, NULL);
727 static void client_common_put_super(struct super_block *sb)
729 struct ll_sb_info *sbi = ll_s2sbi(sb);
734 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
735 obd_disconnect(sbi->ll_dt_exp);
736 sbi->ll_dt_exp = NULL;
738 lprocfs_ll_unregister_mountpoint(sbi);
740 obd_fid_fini(sbi->ll_md_exp->exp_obd);
741 obd_disconnect(sbi->ll_md_exp);
742 sbi->ll_md_exp = NULL;
747 void ll_kill_super(struct super_block *sb)
749 struct ll_sb_info *sbi;
753 if (!(sb->s_flags & MS_ACTIVE))
757 /* we need restore s_dev from changed for clustred NFS before put_super
758 * because new kernels have cached s_dev and change sb->s_dev in
759 * put_super not affected real removing devices */
761 sb->s_dev = sbi->ll_sdev_orig;
762 sbi->ll_umounting = 1;
764 /* wait running statahead threads to quit */
765 while (atomic_read(&sbi->ll_sa_running) > 0) {
766 set_current_state(TASK_UNINTERRUPTIBLE);
767 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
774 static inline int ll_set_opt(const char *opt, char *data, int fl)
776 if (strncmp(opt, data, strlen(opt)) != 0)
782 /* non-client-specific mount options are parsed in lmd_parse */
783 static int ll_options(char *options, int *flags)
786 char *s1 = options, *s2;
792 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
795 CDEBUG(D_SUPER, "next opt=%s\n", s1);
796 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
801 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
806 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
811 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
816 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
821 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
826 tmp = ll_set_opt("context", s1, 1);
829 tmp = ll_set_opt("fscontext", s1, 1);
832 tmp = ll_set_opt("defcontext", s1, 1);
835 tmp = ll_set_opt("rootcontext", s1, 1);
838 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
843 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
849 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
854 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
859 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
864 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
869 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
874 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
879 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
884 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
889 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
894 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
899 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
905 s2 = strchr(s1, ',');
913 void ll_lli_init(struct ll_inode_info *lli)
915 lli->lli_inode_magic = LLI_INODE_MAGIC;
917 spin_lock_init(&lli->lli_lock);
918 lli->lli_posix_acl = NULL;
919 /* Do not set lli_fid, it has been initialized already. */
920 fid_zero(&lli->lli_pfid);
921 lli->lli_mds_read_och = NULL;
922 lli->lli_mds_write_och = NULL;
923 lli->lli_mds_exec_och = NULL;
924 lli->lli_open_fd_read_count = 0;
925 lli->lli_open_fd_write_count = 0;
926 lli->lli_open_fd_exec_count = 0;
927 mutex_init(&lli->lli_och_mutex);
928 spin_lock_init(&lli->lli_agl_lock);
929 spin_lock_init(&lli->lli_layout_lock);
930 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
931 lli->lli_clob = NULL;
933 init_rwsem(&lli->lli_xattrs_list_rwsem);
934 mutex_init(&lli->lli_xattrs_enq_lock);
936 LASSERT(lli->lli_vfs_inode.i_mode != 0);
937 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
938 mutex_init(&lli->lli_readdir_mutex);
939 lli->lli_opendir_key = NULL;
941 spin_lock_init(&lli->lli_sa_lock);
942 lli->lli_opendir_pid = 0;
943 lli->lli_sa_enabled = 0;
944 lli->lli_def_stripe_offset = -1;
946 mutex_init(&lli->lli_size_mutex);
947 lli->lli_symlink_name = NULL;
948 init_rwsem(&lli->lli_trunc_sem);
949 range_lock_tree_init(&lli->lli_write_tree);
950 init_rwsem(&lli->lli_glimpse_sem);
951 lli->lli_glimpse_time = ktime_set(0, 0);
952 INIT_LIST_HEAD(&lli->lli_agl_list);
953 lli->lli_agl_index = 0;
954 lli->lli_async_rc = 0;
956 mutex_init(&lli->lli_layout_mutex);
957 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
960 #ifndef HAVE_SUPER_SETUP_BDI_NAME
962 #define LSI_BDI_INITIALIZED 0x00400000
964 #ifndef HAVE_BDI_CAP_MAP_COPY
965 # define BDI_CAP_MAP_COPY 0
968 #define MAX_STRING_SIZE 128
970 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
972 struct lustre_sb_info *lsi = s2lsi(sb);
973 char buf[MAX_STRING_SIZE];
977 err = bdi_init(&lsi->lsi_bdi);
981 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
982 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
983 lsi->lsi_bdi.name = "lustre";
985 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
987 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
990 sb->s_bdi = &lsi->lsi_bdi;
994 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
996 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
998 struct lustre_profile *lprof = NULL;
999 struct lustre_sb_info *lsi = s2lsi(sb);
1000 struct ll_sb_info *sbi;
1001 char *dt = NULL, *md = NULL;
1002 char *profilenm = get_profile_name(sb);
1003 struct config_llog_instance *cfg;
1004 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1005 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1011 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1017 try_module_get(THIS_MODULE);
1019 /* client additional sb info */
1020 lsi->lsi_llsbi = sbi = ll_init_sbi();
1022 module_put(THIS_MODULE);
1027 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1029 GOTO(out_free, err);
1031 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1033 GOTO(out_free, err);
1035 #ifndef HAVE_DCACHE_LOCK
1036 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1037 sb->s_d_op = &ll_d_ops;
1040 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1041 * so that "llite.*.*" params can be processed correctly. */
1042 if (proc_lustre_fs_root != NULL) {
1043 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1045 CERROR("%s: could not register mountpoint in llite: "
1046 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1051 /* Generate a string unique to this super, in case some joker tries
1052 to mount the same fs at two mount points.
1053 Use the address of the super itself.*/
1054 cfg->cfg_instance = sb;
1055 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1056 cfg->cfg_callback = class_config_llog_handler;
1057 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1058 /* set up client obds */
1059 err = lustre_process_log(sb, profilenm, cfg);
1061 GOTO(out_proc, err);
1063 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1064 lprof = class_get_profile(profilenm);
1065 if (lprof == NULL) {
1066 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1067 " read from the MGS. Does that filesystem "
1068 "exist?\n", profilenm);
1069 GOTO(out_proc, err = -EINVAL);
1071 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1072 lprof->lp_md, lprof->lp_dt);
1074 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1075 OBD_ALLOC(dt, dt_len);
1077 GOTO(out_proc, err = -ENOMEM);
1078 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1080 md_len = strlen(lprof->lp_md) + instlen + 2;
1081 OBD_ALLOC(md, md_len);
1083 GOTO(out_proc, err = -ENOMEM);
1084 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1086 /* connections, registrations, sb setup */
1087 err = client_common_fill_super(sb, md, dt, mnt);
1089 GOTO(out_proc, err);
1091 sbi->ll_client_common_fill_super_succeeded = 1;
1095 lprocfs_ll_unregister_mountpoint(sbi);
1098 OBD_FREE(md, md_len);
1100 OBD_FREE(dt, dt_len);
1102 class_put_profile(lprof);
1105 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1106 LCONSOLE_WARN("Mounted %s\n", profilenm);
1110 } /* ll_fill_super */
1112 void ll_put_super(struct super_block *sb)
1114 struct config_llog_instance cfg, params_cfg;
1115 struct obd_device *obd;
1116 struct lustre_sb_info *lsi = s2lsi(sb);
1117 struct ll_sb_info *sbi = ll_s2sbi(sb);
1118 char *profilenm = get_profile_name(sb);
1120 int next, force = 1, rc = 0;
1123 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1125 cfg.cfg_instance = sb;
1126 lustre_end_log(sb, profilenm, &cfg);
1128 params_cfg.cfg_instance = sb;
1129 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1131 if (sbi->ll_md_exp) {
1132 obd = class_exp2obd(sbi->ll_md_exp);
1134 force = obd->obd_force;
1137 /* Wait for unstable pages to be committed to stable storage */
1139 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1140 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1141 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1145 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1146 if (force == 0 && rc != -EINTR)
1147 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1150 /* We need to set force before the lov_disconnect in
1151 lustre_common_put_super, since l_d cleans up osc's as well. */
1154 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1156 obd->obd_force = force;
1160 if (sbi->ll_client_common_fill_super_succeeded) {
1161 /* Only if client_common_fill_super succeeded */
1162 client_common_put_super(sb);
1166 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1167 class_manual_cleanup(obd);
1170 if (sbi->ll_flags & LL_SBI_VERBOSE)
1171 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1174 class_del_profile(profilenm);
1176 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1177 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1178 bdi_destroy(&lsi->lsi_bdi);
1179 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1184 lsi->lsi_llsbi = NULL;
1186 lustre_common_put_super(sb);
1188 cl_env_cache_purge(~0);
1190 module_put(THIS_MODULE);
1193 } /* client_put_super */
1195 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1197 struct inode *inode = NULL;
1199 /* NOTE: we depend on atomic igrab() -bzzz */
1200 lock_res_and_lock(lock);
1201 if (lock->l_resource->lr_lvb_inode) {
1202 struct ll_inode_info * lli;
1203 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1204 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1205 inode = igrab(lock->l_resource->lr_lvb_inode);
1207 inode = lock->l_resource->lr_lvb_inode;
1208 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1209 D_WARNING, lock, "lr_lvb_inode %p is "
1210 "bogus: magic %08x",
1211 lock->l_resource->lr_lvb_inode,
1212 lli->lli_inode_magic);
1216 unlock_res_and_lock(lock);
1220 void ll_dir_clear_lsm_md(struct inode *inode)
1222 struct ll_inode_info *lli = ll_i2info(inode);
1224 LASSERT(S_ISDIR(inode->i_mode));
1226 if (lli->lli_lsm_md != NULL) {
1227 lmv_free_memmd(lli->lli_lsm_md);
1228 lli->lli_lsm_md = NULL;
1232 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1233 const struct lu_fid *fid,
1234 struct lustre_md *md)
1236 struct ll_sb_info *sbi = ll_s2sbi(sb);
1237 struct mdt_body *body = md->body;
1238 struct inode *inode;
1242 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1243 inode = iget_locked(sb, ino);
1244 if (inode == NULL) {
1245 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1246 ll_get_fsname(sb, NULL, 0), PFID(fid));
1247 RETURN(ERR_PTR(-ENOENT));
1250 if (inode->i_state & I_NEW) {
1251 struct ll_inode_info *lli = ll_i2info(inode);
1252 struct lmv_stripe_md *lsm = md->lmv;
1254 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1255 (body->mbo_mode & S_IFMT);
1256 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1259 LTIME_S(inode->i_mtime) = 0;
1260 LTIME_S(inode->i_atime) = 0;
1261 LTIME_S(inode->i_ctime) = 0;
1264 #ifdef HAVE_BACKING_DEV_INFO
1265 /* initializing backing dev info. */
1266 inode->i_mapping->backing_dev_info =
1267 &s2lsi(inode->i_sb)->lsi_bdi;
1269 inode->i_op = &ll_dir_inode_operations;
1270 inode->i_fop = &ll_dir_operations;
1271 lli->lli_fid = *fid;
1274 LASSERT(lsm != NULL);
1275 /* master object FID */
1276 lli->lli_pfid = body->mbo_fid1;
1277 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1278 lli, PFID(fid), PFID(&lli->lli_pfid));
1279 unlock_new_inode(inode);
1285 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1288 struct lmv_stripe_md *lsm = md->lmv;
1291 LASSERT(lsm != NULL);
1292 /* XXX sigh, this lsm_root initialization should be in
1293 * LMV layer, but it needs ll_iget right now, so we
1294 * put this here right now. */
1295 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1296 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1297 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1298 /* Unfortunately ll_iget will call ll_update_inode,
1299 * where the initialization of slave inode is slightly
1300 * different, so it reset lsm_md to NULL to avoid
1301 * initializing lsm for slave inode. */
1302 /* For migrating inode, master stripe and master object will
1303 * be same, so we only need assign this inode */
1304 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1305 lsm->lsm_md_oinfo[i].lmo_root = inode;
1307 lsm->lsm_md_oinfo[i].lmo_root =
1308 ll_iget_anon_dir(inode->i_sb, fid, md);
1310 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1311 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1313 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1321 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1322 const struct lmv_stripe_md *lsm_md2)
1324 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1325 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1326 lsm_md1->lsm_md_master_mdt_index ==
1327 lsm_md2->lsm_md_master_mdt_index &&
1328 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1329 lsm_md1->lsm_md_layout_version ==
1330 lsm_md2->lsm_md_layout_version &&
1331 strcmp(lsm_md1->lsm_md_pool_name,
1332 lsm_md2->lsm_md_pool_name) == 0;
1335 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1337 struct ll_inode_info *lli = ll_i2info(inode);
1338 struct lmv_stripe_md *lsm = md->lmv;
1342 LASSERT(S_ISDIR(inode->i_mode));
1343 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1344 PFID(ll_inode2fid(inode)));
1346 /* no striped information from request. */
1348 if (lli->lli_lsm_md == NULL) {
1350 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1351 LMV_HASH_FLAG_MIGRATION) {
1352 /* migration is done, the temporay MIGRATE layout has
1354 CDEBUG(D_INODE, DFID" finish migration.\n",
1355 PFID(ll_inode2fid(inode)));
1356 lmv_free_memmd(lli->lli_lsm_md);
1357 lli->lli_lsm_md = NULL;
1360 /* The lustre_md from req does not include stripeEA,
1361 * see ll_md_setattr */
1366 /* set the directory layout */
1367 if (lli->lli_lsm_md == NULL) {
1368 struct cl_attr *attr;
1370 rc = ll_init_lsm_md(inode, md);
1374 /* set md->lmv to NULL, so the following free lustre_md
1375 * will not free this lsm */
1377 lli->lli_lsm_md = lsm;
1379 OBD_ALLOC_PTR(attr);
1383 /* validate the lsm */
1384 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1385 ll_md_blocking_ast);
1391 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1392 md->body->mbo_nlink = attr->cat_nlink;
1393 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1394 md->body->mbo_size = attr->cat_size;
1395 if (md->body->mbo_valid & OBD_MD_FLATIME)
1396 md->body->mbo_atime = attr->cat_atime;
1397 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1398 md->body->mbo_ctime = attr->cat_ctime;
1399 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1400 md->body->mbo_mtime = attr->cat_mtime;
1404 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1405 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1409 /* Compare the old and new stripe information */
1410 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1411 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1414 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1415 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1416 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1417 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1418 inode, lsm, old_lsm,
1419 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1420 lsm->lsm_md_stripe_count,
1421 old_lsm->lsm_md_stripe_count,
1422 lsm->lsm_md_master_mdt_index,
1423 old_lsm->lsm_md_master_mdt_index,
1424 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1425 lsm->lsm_md_layout_version,
1426 old_lsm->lsm_md_layout_version,
1427 lsm->lsm_md_pool_name,
1428 old_lsm->lsm_md_pool_name);
1430 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1431 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1432 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1433 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1436 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1437 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1438 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1439 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1448 void ll_clear_inode(struct inode *inode)
1450 struct ll_inode_info *lli = ll_i2info(inode);
1451 struct ll_sb_info *sbi = ll_i2sbi(inode);
1454 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1455 PFID(ll_inode2fid(inode)), inode);
1457 if (S_ISDIR(inode->i_mode)) {
1458 /* these should have been cleared in ll_file_release */
1459 LASSERT(lli->lli_opendir_key == NULL);
1460 LASSERT(lli->lli_sai == NULL);
1461 LASSERT(lli->lli_opendir_pid == 0);
1464 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1466 LASSERT(!lli->lli_open_fd_write_count);
1467 LASSERT(!lli->lli_open_fd_read_count);
1468 LASSERT(!lli->lli_open_fd_exec_count);
1470 if (lli->lli_mds_write_och)
1471 ll_md_real_close(inode, FMODE_WRITE);
1472 if (lli->lli_mds_exec_och)
1473 ll_md_real_close(inode, FMODE_EXEC);
1474 if (lli->lli_mds_read_och)
1475 ll_md_real_close(inode, FMODE_READ);
1477 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1478 OBD_FREE(lli->lli_symlink_name,
1479 strlen(lli->lli_symlink_name) + 1);
1480 lli->lli_symlink_name = NULL;
1483 ll_xattr_cache_destroy(inode);
1485 #ifdef CONFIG_FS_POSIX_ACL
1486 forget_all_cached_acls(inode);
1487 if (lli->lli_posix_acl) {
1488 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1489 posix_acl_release(lli->lli_posix_acl);
1490 lli->lli_posix_acl = NULL;
1493 lli->lli_inode_magic = LLI_INODE_DEAD;
1495 if (S_ISDIR(inode->i_mode))
1496 ll_dir_clear_lsm_md(inode);
1497 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1498 LASSERT(list_empty(&lli->lli_agl_list));
1501 * XXX This has to be done before lsm is freed below, because
1502 * cl_object still uses inode lsm.
1504 cl_inode_fini(inode);
1509 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1511 struct lustre_md md;
1512 struct inode *inode = dentry->d_inode;
1513 struct ll_sb_info *sbi = ll_i2sbi(inode);
1514 struct ptlrpc_request *request = NULL;
1518 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1519 LUSTRE_OPC_ANY, NULL);
1520 if (IS_ERR(op_data))
1521 RETURN(PTR_ERR(op_data));
1523 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1525 ptlrpc_req_finished(request);
1526 if (rc == -ENOENT) {
1528 /* Unlinked special device node? Or just a race?
1529 * Pretend we done everything. */
1530 if (!S_ISREG(inode->i_mode) &&
1531 !S_ISDIR(inode->i_mode)) {
1532 ia_valid = op_data->op_attr.ia_valid;
1533 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1534 rc = simple_setattr(dentry, &op_data->op_attr);
1535 op_data->op_attr.ia_valid = ia_valid;
1537 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1538 CERROR("md_setattr fails: rc = %d\n", rc);
1543 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1544 sbi->ll_md_exp, &md);
1546 ptlrpc_req_finished(request);
1550 ia_valid = op_data->op_attr.ia_valid;
1551 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1552 * cache is not cleared yet. */
1553 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1554 if (S_ISREG(inode->i_mode))
1556 rc = simple_setattr(dentry, &op_data->op_attr);
1557 if (S_ISREG(inode->i_mode))
1558 inode_unlock(inode);
1559 op_data->op_attr.ia_valid = ia_valid;
1561 rc = ll_update_inode(inode, &md);
1562 ptlrpc_req_finished(request);
1567 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1568 * object(s) determine the file size and mtime. Otherwise, the MDS will
1569 * keep these values until such a time that objects are allocated for it.
1570 * We do the MDS operations first, as it is checking permissions for us.
1571 * We don't to the MDS RPC if there is nothing that we want to store there,
1572 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1573 * going to do an RPC anyways.
1575 * If we are doing a truncate, we will send the mtime and ctime updates
1576 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1577 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1580 * In case of HSMimport, we only set attr on MDS.
1582 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1584 struct inode *inode = dentry->d_inode;
1585 struct ll_inode_info *lli = ll_i2info(inode);
1586 struct md_op_data *op_data = NULL;
1590 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1591 "valid %x, hsm_import %d\n",
1592 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1593 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1596 if (attr->ia_valid & ATTR_SIZE) {
1597 /* Check new size against VFS/VM file size limit and rlimit */
1598 rc = inode_newsize_ok(inode, attr->ia_size);
1602 /* The maximum Lustre file size is variable, based on the
1603 * OST maximum object size and number of stripes. This
1604 * needs another check in addition to the VFS check above. */
1605 if (attr->ia_size > ll_file_maxbytes(inode)) {
1606 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1607 PFID(&lli->lli_fid), attr->ia_size,
1608 ll_file_maxbytes(inode));
1612 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1615 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1616 if (attr->ia_valid & TIMES_SET_FLAGS) {
1617 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1618 !cfs_capable(CFS_CAP_FOWNER))
1622 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1623 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1624 (attr->ia_valid & ATTR_CTIME)) {
1625 attr->ia_ctime = current_time(inode);
1626 attr->ia_valid |= ATTR_CTIME_SET;
1628 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1629 (attr->ia_valid & ATTR_ATIME)) {
1630 attr->ia_atime = current_time(inode);
1631 attr->ia_valid |= ATTR_ATIME_SET;
1633 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1634 (attr->ia_valid & ATTR_MTIME)) {
1635 attr->ia_mtime = current_time(inode);
1636 attr->ia_valid |= ATTR_MTIME_SET;
1639 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1640 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1641 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1642 (s64)ktime_get_real_seconds());
1644 if (S_ISREG(inode->i_mode)) {
1645 if (attr->ia_valid & ATTR_SIZE)
1646 inode_dio_write_done(inode);
1647 inode_unlock(inode);
1650 /* We always do an MDS RPC, even if we're only changing the size;
1651 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1653 OBD_ALLOC_PTR(op_data);
1654 if (op_data == NULL)
1655 GOTO(out, rc = -ENOMEM);
1657 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1658 /* If we are changing file size, file content is
1659 * modified, flag it. */
1660 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1661 op_data->op_bias |= MDS_DATA_MODIFIED;
1662 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1665 op_data->op_attr = *attr;
1667 rc = ll_md_setattr(dentry, op_data);
1671 if (!S_ISREG(inode->i_mode) || hsm_import)
1674 if (attr->ia_valid & (ATTR_SIZE |
1675 ATTR_ATIME | ATTR_ATIME_SET |
1676 ATTR_MTIME | ATTR_MTIME_SET |
1677 ATTR_CTIME | ATTR_CTIME_SET)) {
1678 /* For truncate and utimes sending attributes to OSTs, setting
1679 * mtime/atime to the past will be performed under PW [0:EOF]
1680 * extent lock (new_size:EOF for truncate). It may seem
1681 * excessive to send mtime/atime updates to OSTs when not
1682 * setting times to past, but it is necessary due to possible
1683 * time de-synchronization between MDT inode and OST objects */
1684 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1687 /* If the file was restored, it needs to set dirty flag.
1689 * We've already sent MDS_DATA_MODIFIED flag in
1690 * ll_md_setattr() for truncate. However, the MDT refuses to
1691 * set the HS_DIRTY flag on released files, so we have to set
1692 * it again if the file has been restored. Please check how
1693 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1695 * Please notice that if the file is not released, the previous
1696 * MDS_DATA_MODIFIED has taken effect and usually
1697 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1698 * This way we can save an RPC for common open + trunc
1700 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1701 struct hsm_state_set hss = {
1702 .hss_valid = HSS_SETMASK,
1703 .hss_setmask = HS_DIRTY,
1707 rc2 = ll_hsm_state_set(inode, &hss);
1708 /* truncate and write can happen at the same time, so that
1709 * the file can be set modified even though the file is not
1710 * restored from released state, and ll_hsm_state_set() is
1711 * not applicable for the file, and rc2 < 0 is normal in this
1714 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1715 PFID(ll_inode2fid(inode)), rc2);
1720 if (op_data != NULL)
1721 ll_finish_md_op_data(op_data);
1723 if (S_ISREG(inode->i_mode)) {
1725 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1726 inode_dio_wait(inode);
1727 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1728 * flag. ll_update_inode (called from ll_md_setattr), clears
1729 * inode flags, so there is a gap where S_NOSEC is not set.
1730 * This can cause a writer to take the i_mutex unnecessarily,
1731 * but this is safe to do and should be rare. */
1732 inode_has_no_xattr(inode);
1735 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1736 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1741 int ll_setattr(struct dentry *de, struct iattr *attr)
1743 int mode = de->d_inode->i_mode;
1745 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1746 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1747 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1749 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1750 (ATTR_SIZE|ATTR_MODE)) &&
1751 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1752 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1753 !(attr->ia_mode & S_ISGID))))
1754 attr->ia_valid |= ATTR_FORCE;
1756 if ((attr->ia_valid & ATTR_MODE) &&
1758 !(attr->ia_mode & S_ISUID) &&
1759 !(attr->ia_valid & ATTR_KILL_SUID))
1760 attr->ia_valid |= ATTR_KILL_SUID;
1762 if ((attr->ia_valid & ATTR_MODE) &&
1763 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1764 !(attr->ia_mode & S_ISGID) &&
1765 !(attr->ia_valid & ATTR_KILL_SGID))
1766 attr->ia_valid |= ATTR_KILL_SGID;
1768 /* avoid polluted from ATTR_TIMES_SET,
1769 * projid is not expected to be set here */
1770 attr->ia_valid &= ~MDS_ATTR_PROJID;
1772 return ll_setattr_raw(de, attr, false);
1775 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1776 time64_t max_age, __u32 flags)
1778 struct ll_sb_info *sbi = ll_s2sbi(sb);
1779 struct obd_statfs obd_osfs;
1783 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1785 CERROR("md_statfs fails: rc = %d\n", rc);
1789 osfs->os_type = sb->s_magic;
1791 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1792 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1794 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1795 flags |= OBD_STATFS_NODELAY;
1797 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1799 CERROR("obd_statfs fails: rc = %d\n", rc);
1803 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1804 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1807 osfs->os_bsize = obd_osfs.os_bsize;
1808 osfs->os_blocks = obd_osfs.os_blocks;
1809 osfs->os_bfree = obd_osfs.os_bfree;
1810 osfs->os_bavail = obd_osfs.os_bavail;
1812 /* If we don't have as many objects free on the OST as inodes
1813 * on the MDS, we reduce the total number of inodes to
1814 * compensate, so that the "inodes in use" number is correct.
1816 if (obd_osfs.os_ffree < osfs->os_ffree) {
1817 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1819 osfs->os_ffree = obd_osfs.os_ffree;
1824 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1826 struct super_block *sb = de->d_sb;
1827 struct obd_statfs osfs;
1828 __u64 fsid = huge_encode_dev(sb->s_dev);
1831 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1832 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1834 /* Some amount of caching on the client is allowed */
1835 rc = ll_statfs_internal(sb, &osfs,
1836 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1841 statfs_unpack(sfs, &osfs);
1843 /* We need to downshift for all 32-bit kernels, because we can't
1844 * tell if the kernel is being called via sys_statfs64() or not.
1845 * Stop before overflowing f_bsize - in which case it is better
1846 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1847 if (sizeof(long) < 8) {
1848 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1851 osfs.os_blocks >>= 1;
1852 osfs.os_bfree >>= 1;
1853 osfs.os_bavail >>= 1;
1857 sfs->f_blocks = osfs.os_blocks;
1858 sfs->f_bfree = osfs.os_bfree;
1859 sfs->f_bavail = osfs.os_bavail;
1860 sfs->f_fsid.val[0] = (__u32)fsid;
1861 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1865 void ll_inode_size_lock(struct inode *inode)
1867 struct ll_inode_info *lli;
1869 LASSERT(!S_ISDIR(inode->i_mode));
1871 lli = ll_i2info(inode);
1872 mutex_lock(&lli->lli_size_mutex);
1875 void ll_inode_size_unlock(struct inode *inode)
1877 struct ll_inode_info *lli;
1879 lli = ll_i2info(inode);
1880 mutex_unlock(&lli->lli_size_mutex);
1883 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1885 struct ll_inode_info *lli = ll_i2info(inode);
1886 struct mdt_body *body = md->body;
1887 struct ll_sb_info *sbi = ll_i2sbi(inode);
1890 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1891 rc = cl_file_inode_init(inode, md);
1896 if (S_ISDIR(inode->i_mode)) {
1897 rc = ll_update_lsm_md(inode, md);
1902 #ifdef CONFIG_FS_POSIX_ACL
1903 if (body->mbo_valid & OBD_MD_FLACL) {
1904 spin_lock(&lli->lli_lock);
1905 if (lli->lli_posix_acl)
1906 posix_acl_release(lli->lli_posix_acl);
1907 lli->lli_posix_acl = md->posix_acl;
1908 spin_unlock(&lli->lli_lock);
1911 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1912 sbi->ll_flags & LL_SBI_32BIT_API);
1913 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1915 if (body->mbo_valid & OBD_MD_FLATIME) {
1916 if (body->mbo_atime > LTIME_S(inode->i_atime))
1917 LTIME_S(inode->i_atime) = body->mbo_atime;
1918 lli->lli_atime = body->mbo_atime;
1921 if (body->mbo_valid & OBD_MD_FLMTIME) {
1922 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1923 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1924 "to %llu\n", inode->i_ino,
1925 LTIME_S(inode->i_mtime), body->mbo_mtime);
1926 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1928 lli->lli_mtime = body->mbo_mtime;
1931 if (body->mbo_valid & OBD_MD_FLCTIME) {
1932 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1933 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1934 lli->lli_ctime = body->mbo_ctime;
1937 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1938 if (body->mbo_valid & OBD_MD_FLFLAGS)
1939 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1940 if (body->mbo_valid & OBD_MD_FLMODE)
1941 inode->i_mode = (inode->i_mode & S_IFMT) |
1942 (body->mbo_mode & ~S_IFMT);
1944 if (body->mbo_valid & OBD_MD_FLTYPE)
1945 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1946 (body->mbo_mode & S_IFMT);
1948 LASSERT(inode->i_mode != 0);
1949 if (S_ISREG(inode->i_mode))
1950 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1951 LL_MAX_BLKSIZE_BITS);
1953 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1955 if (body->mbo_valid & OBD_MD_FLUID)
1956 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1957 if (body->mbo_valid & OBD_MD_FLGID)
1958 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1959 if (body->mbo_valid & OBD_MD_FLPROJID)
1960 lli->lli_projid = body->mbo_projid;
1961 if (body->mbo_valid & OBD_MD_FLNLINK)
1962 set_nlink(inode, body->mbo_nlink);
1963 if (body->mbo_valid & OBD_MD_FLRDEV)
1964 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1966 if (body->mbo_valid & OBD_MD_FLID) {
1967 /* FID shouldn't be changed! */
1968 if (fid_is_sane(&lli->lli_fid)) {
1969 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1970 "Trying to change FID "DFID
1971 " to the "DFID", inode "DFID"(%p)\n",
1972 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1973 PFID(ll_inode2fid(inode)), inode);
1975 lli->lli_fid = body->mbo_fid1;
1979 LASSERT(fid_seq(&lli->lli_fid) != 0);
1981 if (body->mbo_valid & OBD_MD_FLSIZE) {
1982 i_size_write(inode, body->mbo_size);
1984 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1985 PFID(ll_inode2fid(inode)),
1986 (unsigned long long)body->mbo_size);
1988 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1989 inode->i_blocks = body->mbo_blocks;
1992 if (body->mbo_valid & OBD_MD_TSTATE) {
1993 /* Set LLIF_FILE_RESTORING if restore ongoing and
1994 * clear it when done to ensure to start again
1995 * glimpsing updated attrs
1997 if (body->mbo_t_state & MS_RESTORE)
1998 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2000 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2006 int ll_read_inode2(struct inode *inode, void *opaque)
2008 struct lustre_md *md = opaque;
2009 struct ll_inode_info *lli = ll_i2info(inode);
2013 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2014 PFID(&lli->lli_fid), inode);
2016 /* Core attributes from the MDS first. This is a new inode, and
2017 * the VFS doesn't zero times in the core inode so we have to do
2018 * it ourselves. They will be overwritten by either MDS or OST
2019 * attributes - we just need to make sure they aren't newer. */
2020 LTIME_S(inode->i_mtime) = 0;
2021 LTIME_S(inode->i_atime) = 0;
2022 LTIME_S(inode->i_ctime) = 0;
2024 rc = ll_update_inode(inode, md);
2028 /* OIDEBUG(inode); */
2030 #ifdef HAVE_BACKING_DEV_INFO
2031 /* initializing backing dev info. */
2032 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2034 if (S_ISREG(inode->i_mode)) {
2035 struct ll_sb_info *sbi = ll_i2sbi(inode);
2036 inode->i_op = &ll_file_inode_operations;
2037 inode->i_fop = sbi->ll_fop;
2038 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2040 } else if (S_ISDIR(inode->i_mode)) {
2041 inode->i_op = &ll_dir_inode_operations;
2042 inode->i_fop = &ll_dir_operations;
2044 } else if (S_ISLNK(inode->i_mode)) {
2045 inode->i_op = &ll_fast_symlink_inode_operations;
2048 inode->i_op = &ll_special_inode_operations;
2050 init_special_inode(inode, inode->i_mode,
2059 void ll_delete_inode(struct inode *inode)
2061 struct ll_inode_info *lli = ll_i2info(inode);
2064 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2065 /* It is last chance to write out dirty pages,
2066 * otherwise we may lose data while umount */
2067 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2069 truncate_inode_pages_final(&inode->i_data);
2071 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2072 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2073 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2075 #ifdef HAVE_SBOPS_EVICT_INODE
2076 ll_clear_inode(inode);
2083 int ll_iocontrol(struct inode *inode, struct file *file,
2084 unsigned int cmd, unsigned long arg)
2086 struct ll_sb_info *sbi = ll_i2sbi(inode);
2087 struct ptlrpc_request *req = NULL;
2092 case FSFILT_IOC_GETFLAGS: {
2093 struct mdt_body *body;
2094 struct md_op_data *op_data;
2096 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2097 0, 0, LUSTRE_OPC_ANY,
2099 if (IS_ERR(op_data))
2100 RETURN(PTR_ERR(op_data));
2102 op_data->op_valid = OBD_MD_FLFLAGS;
2103 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2104 ll_finish_md_op_data(op_data);
2106 CERROR("%s: failure inode "DFID": rc = %d\n",
2107 sbi->ll_md_exp->exp_obd->obd_name,
2108 PFID(ll_inode2fid(inode)), rc);
2112 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2114 flags = body->mbo_flags;
2116 ptlrpc_req_finished(req);
2118 RETURN(put_user(flags, (int __user *)arg));
2120 case FSFILT_IOC_SETFLAGS: {
2122 struct md_op_data *op_data;
2123 struct cl_object *obj;
2125 if (get_user(flags, (int __user *)arg))
2128 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2129 LUSTRE_OPC_ANY, NULL);
2130 if (IS_ERR(op_data))
2131 RETURN(PTR_ERR(op_data));
2133 op_data->op_attr_flags = flags;
2134 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2135 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2136 ll_finish_md_op_data(op_data);
2137 ptlrpc_req_finished(req);
2141 inode->i_flags = ll_ext_to_inode_flags(flags);
2143 obj = ll_i2info(inode)->lli_clob;
2147 OBD_ALLOC_PTR(attr);
2151 attr->ia_valid = ATTR_ATTR_FLAG;
2152 rc = cl_setattr_ost(obj, attr, flags);
2164 int ll_flush_ctx(struct inode *inode)
2166 struct ll_sb_info *sbi = ll_i2sbi(inode);
2168 CDEBUG(D_SEC, "flush context for user %d\n",
2169 from_kuid(&init_user_ns, current_uid()));
2171 obd_set_info_async(NULL, sbi->ll_md_exp,
2172 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2174 obd_set_info_async(NULL, sbi->ll_dt_exp,
2175 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2180 /* umount -f client means force down, don't save state */
2181 void ll_umount_begin(struct super_block *sb)
2183 struct ll_sb_info *sbi = ll_s2sbi(sb);
2184 struct obd_device *obd;
2185 struct obd_ioctl_data *ioc_data;
2186 struct l_wait_info lwi;
2187 wait_queue_head_t waitq;
2190 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2191 sb->s_count, atomic_read(&sb->s_active));
2193 obd = class_exp2obd(sbi->ll_md_exp);
2195 CERROR("Invalid MDC connection handle %#llx\n",
2196 sbi->ll_md_exp->exp_handle.h_cookie);
2202 obd = class_exp2obd(sbi->ll_dt_exp);
2204 CERROR("Invalid LOV connection handle %#llx\n",
2205 sbi->ll_dt_exp->exp_handle.h_cookie);
2211 OBD_ALLOC_PTR(ioc_data);
2213 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2214 sizeof *ioc_data, ioc_data, NULL);
2216 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2217 sizeof *ioc_data, ioc_data, NULL);
2219 OBD_FREE_PTR(ioc_data);
2222 /* Really, we'd like to wait until there are no requests outstanding,
2223 * and then continue. For now, we just periodically checking for vfs
2224 * to decrement mnt_cnt and hope to finish it within 10sec.
2226 init_waitqueue_head(&waitq);
2227 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2228 cfs_time_seconds(1), NULL, NULL);
2229 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2234 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2236 struct ll_sb_info *sbi = ll_s2sbi(sb);
2237 char *profilenm = get_profile_name(sb);
2241 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2242 read_only = *flags & MS_RDONLY;
2243 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2244 sizeof(KEY_READ_ONLY),
2245 KEY_READ_ONLY, sizeof(read_only),
2248 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2249 profilenm, read_only ?
2250 "read-only" : "read-write", err);
2255 sb->s_flags |= MS_RDONLY;
2257 sb->s_flags &= ~MS_RDONLY;
2259 if (sbi->ll_flags & LL_SBI_VERBOSE)
2260 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2261 read_only ? "read-only" : "read-write");
2267 * Cleanup the open handle that is cached on MDT-side.
2269 * For open case, the client side open handling thread may hit error
2270 * after the MDT grant the open. Under such case, the client should
2271 * send close RPC to the MDT as cleanup; otherwise, the open handle
2272 * on the MDT will be leaked there until the client umount or evicted.
2274 * In further, if someone unlinked the file, because the open handle
2275 * holds the reference on such file/object, then it will block the
2276 * subsequent threads that want to locate such object via FID.
2278 * \param[in] sb super block for this file-system
2279 * \param[in] open_req pointer to the original open request
2281 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2283 struct mdt_body *body;
2284 struct md_op_data *op_data;
2285 struct ptlrpc_request *close_req = NULL;
2286 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2289 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2290 OBD_ALLOC_PTR(op_data);
2291 if (op_data == NULL) {
2292 CWARN("%s: cannot allocate op_data to release open handle for "
2294 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2299 op_data->op_fid1 = body->mbo_fid1;
2300 op_data->op_handle = body->mbo_handle;
2301 op_data->op_mod_time = ktime_get_real_seconds();
2302 md_close(exp, op_data, NULL, &close_req);
2303 ptlrpc_req_finished(close_req);
2304 ll_finish_md_op_data(op_data);
2309 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2310 struct super_block *sb, struct lookup_intent *it)
2312 struct ll_sb_info *sbi = NULL;
2313 struct lustre_md md = { NULL };
2317 LASSERT(*inode || sb);
2318 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2319 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2320 sbi->ll_md_exp, &md);
2325 rc = ll_update_inode(*inode, &md);
2329 LASSERT(sb != NULL);
2332 * At this point server returns to client's same fid as client
2333 * generated for creating. So using ->fid1 is okay here.
2335 if (!fid_is_sane(&md.body->mbo_fid1)) {
2336 CERROR("%s: Fid is insane "DFID"\n",
2337 ll_get_fsname(sb, NULL, 0),
2338 PFID(&md.body->mbo_fid1));
2339 GOTO(out, rc = -EINVAL);
2342 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2343 sbi->ll_flags & LL_SBI_32BIT_API),
2345 if (IS_ERR(*inode)) {
2346 #ifdef CONFIG_FS_POSIX_ACL
2348 posix_acl_release(md.posix_acl);
2349 md.posix_acl = NULL;
2352 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2354 CERROR("new_inode -fatal: rc %d\n", rc);
2359 /* Handling piggyback layout lock.
2360 * Layout lock can be piggybacked by getattr and open request.
2361 * The lsm can be applied to inode only if it comes with a layout lock
2362 * otherwise correct layout may be overwritten, for example:
2363 * 1. proc1: mdt returns a lsm but not granting layout
2364 * 2. layout was changed by another client
2365 * 3. proc2: refresh layout and layout lock granted
2366 * 4. proc1: to apply a stale layout */
2367 if (it != NULL && it->it_lock_mode != 0) {
2368 struct lustre_handle lockh;
2369 struct ldlm_lock *lock;
2371 lockh.cookie = it->it_lock_handle;
2372 lock = ldlm_handle2lock(&lockh);
2373 LASSERT(lock != NULL);
2374 if (ldlm_has_layout(lock)) {
2375 struct cl_object_conf conf;
2377 memset(&conf, 0, sizeof(conf));
2378 conf.coc_opc = OBJECT_CONF_SET;
2379 conf.coc_inode = *inode;
2380 conf.coc_lock = lock;
2381 conf.u.coc_layout = md.layout;
2382 (void)ll_layout_conf(*inode, &conf);
2384 LDLM_LOCK_PUT(lock);
2390 md_free_lustre_md(sbi->ll_md_exp, &md);
2393 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2394 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2399 int ll_obd_statfs(struct inode *inode, void __user *arg)
2401 struct ll_sb_info *sbi = NULL;
2402 struct obd_export *exp;
2404 struct obd_ioctl_data *data = NULL;
2408 if (!inode || !(sbi = ll_i2sbi(inode)))
2409 GOTO(out_statfs, rc = -EINVAL);
2411 rc = obd_ioctl_getdata(&buf, &len, arg);
2413 GOTO(out_statfs, rc);
2416 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2417 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2418 GOTO(out_statfs, rc = -EINVAL);
2420 if (data->ioc_inllen1 != sizeof(__u32) ||
2421 data->ioc_inllen2 != sizeof(__u32) ||
2422 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2423 data->ioc_plen2 != sizeof(struct obd_uuid))
2424 GOTO(out_statfs, rc = -EINVAL);
2426 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2427 if (type & LL_STATFS_LMV)
2428 exp = sbi->ll_md_exp;
2429 else if (type & LL_STATFS_LOV)
2430 exp = sbi->ll_dt_exp;
2432 GOTO(out_statfs, rc = -ENODEV);
2434 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2436 GOTO(out_statfs, rc);
2438 OBD_FREE_LARGE(buf, len);
2442 int ll_process_config(struct lustre_cfg *lcfg)
2444 struct super_block *sb;
2449 /* The instance name contains the sb: lustre-client-aacfe000 */
2450 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2451 if (!ptr || !*(++ptr))
2453 if (sscanf(ptr, "%lx", &x) != 1)
2455 sb = (struct super_block *)x;
2456 /* This better be a real Lustre superblock! */
2457 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2459 /* Note we have not called client_common_fill_super yet, so
2460 proc fns must be able to handle that! */
2461 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2468 /* this function prepares md_op_data hint for passing it down to MD stack. */
2469 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2470 struct inode *i1, struct inode *i2,
2471 const char *name, size_t namelen,
2472 __u32 mode, __u32 opc, void *data)
2474 LASSERT(i1 != NULL);
2477 /* Do not reuse namelen for something else. */
2479 return ERR_PTR(-EINVAL);
2481 if (namelen > ll_i2sbi(i1)->ll_namelen)
2482 return ERR_PTR(-ENAMETOOLONG);
2484 if (!lu_name_is_valid_2(name, namelen))
2485 return ERR_PTR(-EINVAL);
2488 if (op_data == NULL)
2489 OBD_ALLOC_PTR(op_data);
2491 if (op_data == NULL)
2492 return ERR_PTR(-ENOMEM);
2494 ll_i2gids(op_data->op_suppgids, i1, i2);
2495 op_data->op_fid1 = *ll_inode2fid(i1);
2496 op_data->op_default_stripe_offset = -1;
2497 if (S_ISDIR(i1->i_mode)) {
2498 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2499 if (opc == LUSTRE_OPC_MKDIR)
2500 op_data->op_default_stripe_offset =
2501 ll_i2info(i1)->lli_def_stripe_offset;
2505 op_data->op_fid2 = *ll_inode2fid(i2);
2506 if (S_ISDIR(i2->i_mode))
2507 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2509 fid_zero(&op_data->op_fid2);
2512 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2513 op_data->op_cli_flags |= CLI_HASH64;
2515 if (ll_need_32bit_api(ll_i2sbi(i1)))
2516 op_data->op_cli_flags |= CLI_API32;
2518 op_data->op_name = name;
2519 op_data->op_namelen = namelen;
2520 op_data->op_mode = mode;
2521 op_data->op_mod_time = ktime_get_real_seconds();
2522 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2523 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2524 op_data->op_cap = cfs_curproc_cap_pack();
2525 op_data->op_mds = 0;
2526 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2527 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2528 op_data->op_bias |= MDS_CREATE_VOLATILE;
2530 op_data->op_data = data;
2535 void ll_finish_md_op_data(struct md_op_data *op_data)
2537 security_release_secctx(op_data->op_file_secctx,
2538 op_data->op_file_secctx_size);
2539 OBD_FREE_PTR(op_data);
2542 #ifdef HAVE_SUPEROPS_USE_DENTRY
2543 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2545 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2548 struct ll_sb_info *sbi;
2550 #ifdef HAVE_SUPEROPS_USE_DENTRY
2551 LASSERT((seq != NULL) && (dentry != NULL));
2552 sbi = ll_s2sbi(dentry->d_sb);
2554 LASSERT((seq != NULL) && (vfs != NULL));
2555 sbi = ll_s2sbi(vfs->mnt_sb);
2558 if (sbi->ll_flags & LL_SBI_NOLCK)
2559 seq_puts(seq, ",nolock");
2561 if (sbi->ll_flags & LL_SBI_FLOCK)
2562 seq_puts(seq, ",flock");
2564 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2565 seq_puts(seq, ",localflock");
2567 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2568 seq_puts(seq, ",user_xattr");
2570 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2571 seq_puts(seq, ",lazystatfs");
2573 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2574 seq_puts(seq, ",user_fid2path");
2576 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2577 seq_puts(seq, ",always_ping");
2583 * Get obd name by cmd, and copy out to user space
2585 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2587 struct ll_sb_info *sbi = ll_i2sbi(inode);
2588 struct obd_device *obd;
2591 if (cmd == OBD_IOC_GETDTNAME)
2592 obd = class_exp2obd(sbi->ll_dt_exp);
2593 else if (cmd == OBD_IOC_GETMDNAME)
2594 obd = class_exp2obd(sbi->ll_md_exp);
2601 if (copy_to_user((void __user *)arg, obd->obd_name,
2602 strlen(obd->obd_name) + 1))
2609 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2610 * fsname will be returned in this buffer; otherwise, a static buffer will be
2611 * used to store the fsname and returned to caller.
2613 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2615 static char fsname_static[MTI_NAME_MAXLEN];
2616 struct lustre_sb_info *lsi = s2lsi(sb);
2621 /* this means the caller wants to use static buffer
2622 * and it doesn't care about race. Usually this is
2623 * in error reporting path */
2624 buf = fsname_static;
2625 buflen = sizeof(fsname_static);
2628 len = strlen(lsi->lsi_lmd->lmd_profile);
2629 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2630 if (ptr && (strcmp(ptr, "-client") == 0))
2633 if (unlikely(len >= buflen))
2635 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2641 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2648 p.mnt = current->fs->root.mnt;
2650 path = d_path(&p, buf, bufsize);
2655 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2657 char *buf, *path = NULL;
2658 struct dentry *dentry = NULL;
2659 struct inode *inode = page->mapping->host;
2661 /* this can be called inside spin lock so use GFP_ATOMIC. */
2662 buf = (char *)__get_free_page(GFP_ATOMIC);
2664 dentry = d_find_alias(page->mapping->host);
2666 path = ll_d_path(dentry, buf, PAGE_SIZE);
2670 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2671 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2672 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2673 PFID(ll_inode2fid(inode)),
2674 (path && !IS_ERR(path)) ? path : "", ioret);
2680 free_page((unsigned long)buf);
2683 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2684 struct lov_user_md **kbuf)
2686 struct lov_user_md lum;
2690 if (copy_from_user(&lum, md, sizeof(lum)))
2693 lum_size = ll_lov_user_md_size(&lum);
2697 OBD_ALLOC(*kbuf, lum_size);
2701 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2702 OBD_FREE(*kbuf, lum_size);
2710 * Compute llite root squash state after a change of root squash
2711 * configuration setting or add/remove of a lnet nid
2713 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2715 struct root_squash_info *squash = &sbi->ll_squash;
2718 struct lnet_process_id id;
2720 /* Update norootsquash flag */
2721 down_write(&squash->rsi_sem);
2722 if (list_empty(&squash->rsi_nosquash_nids))
2723 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2725 /* Do not apply root squash as soon as one of our NIDs is
2726 * in the nosquash_nids list */
2729 while (LNetGetId(i++, &id) != -ENOENT) {
2730 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2732 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2738 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2740 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2742 up_write(&squash->rsi_sem);
2746 * Parse linkea content to extract information about a given hardlink
2748 * \param[in] ldata - Initialized linkea data
2749 * \param[in] linkno - Link identifier
2750 * \param[out] parent_fid - The entry's parent FID
2751 * \param[out] ln - Entry name destination buffer
2753 * \retval 0 on success
2754 * \retval Appropriate negative error code on failure
2756 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2757 struct lu_fid *parent_fid, struct lu_name *ln)
2763 rc = linkea_init_with_rec(ldata);
2767 if (linkno >= ldata->ld_leh->leh_reccount)
2768 /* beyond last link */
2771 linkea_first_entry(ldata);
2772 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2773 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2778 linkea_next_entry(ldata);
2788 * Get parent FID and name of an identified link. Operation is performed for
2789 * a given link number, letting the caller iterate over linkno to list one or
2790 * all links of an entry.
2792 * \param[in] file - File descriptor against which to perform the operation
2793 * \param[in,out] arg - User-filled structure containing the linkno to operate
2794 * on and the available size. It is eventually filled with
2795 * the requested information or left untouched on error
2797 * \retval - 0 on success
2798 * \retval - Appropriate negative error code on failure
2800 int ll_getparent(struct file *file, struct getparent __user *arg)
2802 struct inode *inode = file_inode(file);
2803 struct linkea_data *ldata;
2804 struct lu_buf buf = LU_BUF_NULL;
2806 struct lu_fid parent_fid;
2813 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2814 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2817 if (get_user(name_size, &arg->gp_name_size))
2820 if (get_user(linkno, &arg->gp_linkno))
2823 if (name_size > PATH_MAX)
2826 OBD_ALLOC(ldata, sizeof(*ldata));
2830 rc = linkea_data_new(ldata, &buf);
2832 GOTO(ldata_free, rc);
2834 #ifdef HAVE_XATTR_HANDLER_FLAGS
2835 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2836 buf.lb_len, OBD_MD_FLXATTR);
2838 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2840 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2844 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2848 if (ln.ln_namelen >= name_size)
2849 GOTO(lb_free, rc = -EOVERFLOW);
2851 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2852 GOTO(lb_free, rc = -EFAULT);
2854 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2855 GOTO(lb_free, rc = -EFAULT);
2857 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2858 GOTO(lb_free, rc = -EFAULT);
2863 OBD_FREE(ldata, sizeof(*ldata));