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;
133 sbi->ll_flags |= LL_SBI_TINY_WRITE;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* pass client page size via ocd_grant_blkbits, the server should report
200 * back its backend blocksize for grant calculation purpose */
201 data->ocd_grant_blkbits = PAGE_SHIFT;
203 /* indicate MDT features supported by this client */
204 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
205 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
206 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
207 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
208 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
209 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
210 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
211 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
212 OBD_CONNECT_64BITHASH |
213 OBD_CONNECT_EINPROGRESS |
214 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
215 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
216 OBD_CONNECT_MAX_EASIZE |
217 OBD_CONNECT_FLOCK_DEAD |
218 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
219 OBD_CONNECT_OPEN_BY_FID |
220 OBD_CONNECT_DIR_STRIPE |
221 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
222 OBD_CONNECT_SUBTREE |
223 OBD_CONNECT_MULTIMODRPCS |
224 OBD_CONNECT_GRANT_PARAM |
225 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
227 data->ocd_connect_flags2 = OBD_CONNECT2_FLR;
229 #ifdef HAVE_LRU_RESIZE_SUPPORT
230 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
231 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
233 #ifdef CONFIG_FS_POSIX_ACL
234 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
235 OBD_CONNECT_LARGE_ACL;
238 data->ocd_cksum_types = cksum_types_supported_client();
240 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
241 /* flag mdc connection as lightweight, only used for test
242 * purpose, use with care */
243 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
245 data->ocd_ibits_known = MDS_INODELOCK_FULL;
246 data->ocd_version = LUSTRE_VERSION_CODE;
248 if (sb->s_flags & MS_RDONLY)
249 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
250 if (sbi->ll_flags & LL_SBI_USER_XATTR)
251 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
254 /* Setting this indicates we correctly support S_NOSEC (See kernel
255 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
257 sb->s_flags |= MS_NOSEC;
260 if (sbi->ll_flags & LL_SBI_FLOCK)
261 sbi->ll_fop = &ll_file_operations_flock;
262 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
263 sbi->ll_fop = &ll_file_operations;
265 sbi->ll_fop = &ll_file_operations_noflock;
267 /* always ping even if server suppress_pings */
268 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
269 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
271 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
272 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
273 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
275 data->ocd_brw_size = MD_MAX_BRW_SIZE;
277 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
279 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
280 "recovery, of which this client is not a "
281 "part. Please wait for recovery to complete,"
282 " abort, or time out.\n", md);
285 CERROR("cannot connect to %s: rc = %d\n", md, err);
289 sbi->ll_md_exp->exp_connect_data = *data;
291 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
292 LUSTRE_SEQ_METADATA);
294 CERROR("%s: Can't init metadata layer FID infrastructure, "
295 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
299 /* For mount, we only need fs info from MDT0, and also in DNE, it
300 * can make sure the client can be mounted as long as MDT0 is
302 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
303 ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
304 OBD_STATFS_FOR_MDT0);
306 GOTO(out_md_fid, err);
308 /* This needs to be after statfs to ensure connect has finished.
309 * Note that "data" does NOT contain the valid connect reply.
310 * If connecting to a 1.8 server there will be no LMV device, so
311 * we can access the MDC export directly and exp_connect_flags will
312 * be non-zero, but if accessing an upgraded 2.1 server it will
313 * have the correct flags filled in.
314 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
315 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
316 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
317 valid != CLIENT_CONNECT_MDT_REQD) {
320 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
321 obd_connect_flags2str(buf, PAGE_SIZE,
322 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
323 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
324 "feature(s) needed for correct operation "
325 "of this client (%s). Please upgrade "
326 "server or downgrade client.\n",
327 sbi->ll_md_exp->exp_obd->obd_name, buf);
328 OBD_FREE(buf, PAGE_SIZE);
329 GOTO(out_md_fid, err = -EPROTO);
332 size = sizeof(*data);
333 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
334 KEY_CONN_DATA, &size, data);
336 CERROR("%s: Get connect data failed: rc = %d\n",
337 sbi->ll_md_exp->exp_obd->obd_name, err);
338 GOTO(out_md_fid, err);
341 LASSERT(osfs->os_bsize);
342 sb->s_blocksize = osfs->os_bsize;
343 sb->s_blocksize_bits = log2(osfs->os_bsize);
344 sb->s_magic = LL_SUPER_MAGIC;
345 sb->s_maxbytes = MAX_LFS_FILESIZE;
346 sbi->ll_namelen = osfs->os_namelen;
347 sbi->ll_mnt.mnt = current->fs->root.mnt;
349 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
350 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
351 LCONSOLE_INFO("Disabling user_xattr feature because "
352 "it is not supported on the server\n");
353 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
356 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
358 sb->s_flags |= MS_POSIXACL;
360 sbi->ll_flags |= LL_SBI_ACL;
362 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
364 sb->s_flags &= ~MS_POSIXACL;
366 sbi->ll_flags &= ~LL_SBI_ACL;
369 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
370 sbi->ll_flags |= LL_SBI_64BIT_HASH;
372 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
373 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
375 if (obd_connect_has_secctx(data))
376 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
378 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
379 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
380 LCONSOLE_INFO("%s: disabling xattr cache due to "
381 "unknown maximum xattr size.\n", dt);
382 } else if (!sbi->ll_xattr_cache_set) {
383 /* If xattr_cache is already set (no matter 0 or 1)
384 * during processing llog, it won't be enabled here. */
385 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
386 sbi->ll_xattr_cache_enabled = 1;
390 obd = class_name2obd(dt);
392 CERROR("DT %s: not setup or attached\n", dt);
393 GOTO(out_md_fid, err = -ENODEV);
396 /* pass client page size via ocd_grant_blkbits, the server should report
397 * back its backend blocksize for grant calculation purpose */
398 data->ocd_grant_blkbits = PAGE_SHIFT;
400 /* indicate OST features supported by this client */
401 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
402 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
403 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
404 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
405 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
406 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
407 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
408 OBD_CONNECT_EINPROGRESS |
409 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
410 OBD_CONNECT_LAYOUTLOCK |
411 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
412 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
415 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
416 * can interoperate with an older version of lockahead which was released prior
417 * to landing in master. This support will be dropped when 2.13 development
418 * starts. At the point, we should not just drop the connect flag (below), we
419 * should also remove the support in the code.
421 * Removing it means a few things:
422 * 1. Remove this section here
423 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
424 * 3. Remove function exp_connect_lockahead_old
425 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
427 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
428 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
431 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
433 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
434 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
436 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
437 * disabled by default, because it can still be enabled on the
438 * fly via /proc. As a consequence, we still need to come to an
439 * agreement on the supported algorithms at connect time */
440 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
442 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
443 data->ocd_cksum_types = OBD_CKSUM_ADLER;
445 data->ocd_cksum_types = cksum_types_supported_client();
447 #ifdef HAVE_LRU_RESIZE_SUPPORT
448 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
450 /* always ping even if server suppress_pings */
451 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
452 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
454 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
455 "ocd_grant: %d\n", data->ocd_connect_flags,
456 data->ocd_version, data->ocd_grant);
458 obd->obd_upcall.onu_owner = &sbi->ll_lco;
459 obd->obd_upcall.onu_upcall = cl_ocd_update;
461 data->ocd_brw_size = DT_MAX_BRW_SIZE;
463 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
466 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
467 "recovery, of which this client is not a "
468 "part. Please wait for recovery to "
469 "complete, abort, or time out.\n", dt);
472 CERROR("%s: Cannot connect to %s: rc = %d\n",
473 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
477 sbi->ll_dt_exp->exp_connect_data = *data;
479 /* Don't change value if it was specified in the config log */
480 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
481 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
482 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
483 (data->ocd_brw_size >> PAGE_SHIFT));
485 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
486 LUSTRE_SEQ_METADATA);
488 CERROR("%s: Can't init data layer FID infrastructure, "
489 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
493 mutex_lock(&sbi->ll_lco.lco_lock);
494 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
495 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
496 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
497 mutex_unlock(&sbi->ll_lco.lco_lock);
499 fid_zero(&sbi->ll_root_fid);
500 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
503 CERROR("cannot mds_connect: rc = %d\n", err);
504 GOTO(out_lock_cn_cb, err);
506 if (!fid_is_sane(&sbi->ll_root_fid)) {
507 CERROR("%s: Invalid root fid "DFID" during mount\n",
508 sbi->ll_md_exp->exp_obd->obd_name,
509 PFID(&sbi->ll_root_fid));
510 GOTO(out_lock_cn_cb, err = -EINVAL);
512 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
514 sb->s_op = &lustre_super_operations;
515 #ifdef HAVE_XATTR_HANDLER_FLAGS
516 sb->s_xattr = ll_xattr_handlers;
518 #if THREAD_SIZE >= 8192 /*b=17630*/
519 sb->s_export_op = &lustre_export_operations;
523 * XXX: move this to after cbd setup? */
524 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
525 if (sbi->ll_flags & LL_SBI_ACL)
526 valid |= OBD_MD_FLACL;
528 OBD_ALLOC_PTR(op_data);
530 GOTO(out_lock_cn_cb, err = -ENOMEM);
532 op_data->op_fid1 = sbi->ll_root_fid;
533 op_data->op_mode = 0;
534 op_data->op_valid = valid;
536 err = md_getattr(sbi->ll_md_exp, op_data, &request);
538 OBD_FREE_PTR(op_data);
540 CERROR("%s: md_getattr failed for root: rc = %d\n",
541 sbi->ll_md_exp->exp_obd->obd_name, err);
542 GOTO(out_lock_cn_cb, err);
545 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
546 sbi->ll_md_exp, &lmd);
548 CERROR("failed to understand root inode md: rc = %d\n", err);
549 ptlrpc_req_finished(request);
550 GOTO(out_lock_cn_cb, err);
553 LASSERT(fid_is_sane(&sbi->ll_root_fid));
554 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
555 sbi->ll_flags & LL_SBI_32BIT_API),
557 md_free_lustre_md(sbi->ll_md_exp, &lmd);
558 ptlrpc_req_finished(request);
561 #ifdef CONFIG_FS_POSIX_ACL
563 posix_acl_release(lmd.posix_acl);
564 lmd.posix_acl = NULL;
567 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
569 CERROR("lustre_lite: bad iget4 for root\n");
573 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
574 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
575 KEY_CHECKSUM, sizeof(checksum), &checksum,
578 CERROR("%s: Set checksum failed: rc = %d\n",
579 sbi->ll_dt_exp->exp_obd->obd_name, err);
584 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
585 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
586 sbi->ll_cache, NULL);
588 CERROR("%s: Set cache_set failed: rc = %d\n",
589 sbi->ll_dt_exp->exp_obd->obd_name, err);
593 sb->s_root = d_make_root(root);
594 if (sb->s_root == NULL) {
595 CERROR("%s: can't make root dentry\n",
596 ll_get_fsname(sb, NULL, 0));
597 GOTO(out_root, err = -ENOMEM);
599 #ifdef HAVE_DCACHE_LOCK
600 sb->s_root->d_op = &ll_d_ops;
603 sbi->ll_sdev_orig = sb->s_dev;
605 /* We set sb->s_dev equal on all lustre clients in order to support
606 * NFS export clustering. NFSD requires that the FSID be the same
608 /* s_dev is also used in lt_compare() to compare two fs, but that is
609 * only a node-local comparison. */
610 uuid = obd_get_uuid(sbi->ll_md_exp);
612 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
619 if (sbi->ll_proc_root != NULL) {
620 err = lprocfs_ll_register_obd(sb, dt);
622 CERROR("%s: could not register %s in llite: rc = %d\n",
623 dt, ll_get_fsname(sb, NULL, 0), err);
626 err = lprocfs_ll_register_obd(sb, md);
628 CERROR("%s: could not register %s in llite: rc = %d\n",
629 md, ll_get_fsname(sb, NULL, 0), err);
639 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
641 obd_disconnect(sbi->ll_dt_exp);
642 sbi->ll_dt_exp = NULL;
644 obd_fid_fini(sbi->ll_md_exp->exp_obd);
646 obd_disconnect(sbi->ll_md_exp);
647 sbi->ll_md_exp = NULL;
656 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
660 size = sizeof(*lmmsize);
661 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
662 KEY_MAX_EASIZE, &size, lmmsize);
664 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
665 sbi->ll_dt_exp->exp_obd->obd_name, rc);
670 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
671 KEY_MAX_EASIZE, &size, lmmsize);
673 CERROR("Get max mdsize error rc %d\n", rc);
679 * Get the value of the default_easize parameter.
681 * \see client_obd::cl_default_mds_easize
683 * \param[in] sbi superblock info for this filesystem
684 * \param[out] lmmsize pointer to storage location for value
686 * \retval 0 on success
687 * \retval negative negated errno on failure
689 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
694 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
695 KEY_DEFAULT_EASIZE, &size, lmmsize);
697 CERROR("Get default mdsize error rc %d\n", rc);
703 * Set the default_easize parameter to the given value.
705 * \see client_obd::cl_default_mds_easize
707 * \param[in] sbi superblock info for this filesystem
708 * \param[in] lmmsize the size to set
710 * \retval 0 on success
711 * \retval negative negated errno on failure
713 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
717 if (lmmsize < sizeof(struct lov_mds_md) ||
718 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
721 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
722 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
723 sizeof(int), &lmmsize, NULL);
728 static void client_common_put_super(struct super_block *sb)
730 struct ll_sb_info *sbi = ll_s2sbi(sb);
735 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
736 obd_disconnect(sbi->ll_dt_exp);
737 sbi->ll_dt_exp = NULL;
739 lprocfs_ll_unregister_mountpoint(sbi);
741 obd_fid_fini(sbi->ll_md_exp->exp_obd);
742 obd_disconnect(sbi->ll_md_exp);
743 sbi->ll_md_exp = NULL;
748 void ll_kill_super(struct super_block *sb)
750 struct ll_sb_info *sbi;
754 if (!(sb->s_flags & MS_ACTIVE))
758 /* we need restore s_dev from changed for clustred NFS before put_super
759 * because new kernels have cached s_dev and change sb->s_dev in
760 * put_super not affected real removing devices */
762 sb->s_dev = sbi->ll_sdev_orig;
763 sbi->ll_umounting = 1;
765 /* wait running statahead threads to quit */
766 while (atomic_read(&sbi->ll_sa_running) > 0) {
767 set_current_state(TASK_UNINTERRUPTIBLE);
768 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
775 static inline int ll_set_opt(const char *opt, char *data, int fl)
777 if (strncmp(opt, data, strlen(opt)) != 0)
783 /* non-client-specific mount options are parsed in lmd_parse */
784 static int ll_options(char *options, int *flags)
787 char *s1 = options, *s2;
793 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
796 CDEBUG(D_SUPER, "next opt=%s\n", s1);
797 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
802 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
807 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
812 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
817 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
822 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
827 tmp = ll_set_opt("context", s1, 1);
830 tmp = ll_set_opt("fscontext", s1, 1);
833 tmp = ll_set_opt("defcontext", s1, 1);
836 tmp = ll_set_opt("rootcontext", s1, 1);
839 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
844 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
850 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
855 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
860 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
865 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
870 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
875 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
880 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
885 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
890 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
895 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
900 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
906 s2 = strchr(s1, ',');
914 void ll_lli_init(struct ll_inode_info *lli)
916 lli->lli_inode_magic = LLI_INODE_MAGIC;
918 spin_lock_init(&lli->lli_lock);
919 lli->lli_posix_acl = NULL;
920 /* Do not set lli_fid, it has been initialized already. */
921 fid_zero(&lli->lli_pfid);
922 lli->lli_mds_read_och = NULL;
923 lli->lli_mds_write_och = NULL;
924 lli->lli_mds_exec_och = NULL;
925 lli->lli_open_fd_read_count = 0;
926 lli->lli_open_fd_write_count = 0;
927 lli->lli_open_fd_exec_count = 0;
928 mutex_init(&lli->lli_och_mutex);
929 spin_lock_init(&lli->lli_agl_lock);
930 spin_lock_init(&lli->lli_layout_lock);
931 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
932 lli->lli_clob = NULL;
934 init_rwsem(&lli->lli_xattrs_list_rwsem);
935 mutex_init(&lli->lli_xattrs_enq_lock);
937 LASSERT(lli->lli_vfs_inode.i_mode != 0);
938 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
939 mutex_init(&lli->lli_readdir_mutex);
940 lli->lli_opendir_key = NULL;
942 spin_lock_init(&lli->lli_sa_lock);
943 lli->lli_opendir_pid = 0;
944 lli->lli_sa_enabled = 0;
945 lli->lli_def_stripe_offset = -1;
947 mutex_init(&lli->lli_size_mutex);
948 lli->lli_symlink_name = NULL;
949 init_rwsem(&lli->lli_trunc_sem);
950 range_lock_tree_init(&lli->lli_write_tree);
951 init_rwsem(&lli->lli_glimpse_sem);
952 lli->lli_glimpse_time = ktime_set(0, 0);
953 INIT_LIST_HEAD(&lli->lli_agl_list);
954 lli->lli_agl_index = 0;
955 lli->lli_async_rc = 0;
957 mutex_init(&lli->lli_layout_mutex);
958 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
961 #ifndef HAVE_SUPER_SETUP_BDI_NAME
963 #define LSI_BDI_INITIALIZED 0x00400000
965 #ifndef HAVE_BDI_CAP_MAP_COPY
966 # define BDI_CAP_MAP_COPY 0
969 #define MAX_STRING_SIZE 128
971 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
973 struct lustre_sb_info *lsi = s2lsi(sb);
974 char buf[MAX_STRING_SIZE];
978 err = bdi_init(&lsi->lsi_bdi);
982 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
983 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
984 lsi->lsi_bdi.name = "lustre";
986 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
988 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
991 sb->s_bdi = &lsi->lsi_bdi;
995 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
997 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
999 struct lustre_profile *lprof = NULL;
1000 struct lustre_sb_info *lsi = s2lsi(sb);
1001 struct ll_sb_info *sbi = NULL;
1002 char *dt = NULL, *md = NULL;
1003 char *profilenm = get_profile_name(sb);
1004 struct config_llog_instance *cfg;
1005 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1006 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1012 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1014 try_module_get(THIS_MODULE);
1018 GOTO(out_free, err = -ENOMEM);
1020 /* client additional sb info */
1021 lsi->lsi_llsbi = sbi = ll_init_sbi();
1023 GOTO(out_free, err = -ENOMEM);
1025 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1027 GOTO(out_free, err);
1029 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1031 GOTO(out_free, err);
1033 #ifndef HAVE_DCACHE_LOCK
1034 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1035 sb->s_d_op = &ll_d_ops;
1038 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1039 * so that "llite.*.*" params can be processed correctly. */
1040 if (proc_lustre_fs_root != NULL) {
1041 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1043 CERROR("%s: could not register mountpoint in llite: "
1044 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1049 /* Generate a string unique to this super, in case some joker tries
1050 to mount the same fs at two mount points.
1051 Use the address of the super itself.*/
1052 cfg->cfg_instance = sb;
1053 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1054 cfg->cfg_callback = class_config_llog_handler;
1055 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1056 /* set up client obds */
1057 err = lustre_process_log(sb, profilenm, cfg);
1059 GOTO(out_proc, err);
1061 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1062 lprof = class_get_profile(profilenm);
1063 if (lprof == NULL) {
1064 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1065 " read from the MGS. Does that filesystem "
1066 "exist?\n", profilenm);
1067 GOTO(out_proc, err = -EINVAL);
1069 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1070 lprof->lp_md, lprof->lp_dt);
1072 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1073 OBD_ALLOC(dt, dt_len);
1075 GOTO(out_proc, err = -ENOMEM);
1076 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1078 md_len = strlen(lprof->lp_md) + instlen + 2;
1079 OBD_ALLOC(md, md_len);
1081 GOTO(out_proc, err = -ENOMEM);
1082 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1084 /* connections, registrations, sb setup */
1085 err = client_common_fill_super(sb, md, dt, mnt);
1087 GOTO(out_proc, err);
1089 sbi->ll_client_common_fill_super_succeeded = 1;
1093 lprocfs_ll_unregister_mountpoint(sbi);
1096 OBD_FREE(md, md_len);
1098 OBD_FREE(dt, dt_len);
1100 class_put_profile(lprof);
1105 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1106 LCONSOLE_WARN("Mounted %s\n", profilenm);
1108 } /* ll_fill_super */
1110 void ll_put_super(struct super_block *sb)
1112 struct config_llog_instance cfg, params_cfg;
1113 struct obd_device *obd;
1114 struct lustre_sb_info *lsi = s2lsi(sb);
1115 struct ll_sb_info *sbi = ll_s2sbi(sb);
1116 char *profilenm = get_profile_name(sb);
1118 int next, force = 1, rc = 0;
1122 GOTO(out_no_sbi, 0);
1124 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1126 cfg.cfg_instance = sb;
1127 lustre_end_log(sb, profilenm, &cfg);
1129 params_cfg.cfg_instance = sb;
1130 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1132 if (sbi->ll_md_exp) {
1133 obd = class_exp2obd(sbi->ll_md_exp);
1135 force = obd->obd_force;
1138 /* Wait for unstable pages to be committed to stable storage */
1140 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1141 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1142 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1146 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1147 if (force == 0 && rc != -EINTR)
1148 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1151 /* We need to set force before the lov_disconnect in
1152 lustre_common_put_super, since l_d cleans up osc's as well. */
1155 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1157 obd->obd_force = force;
1161 if (sbi->ll_client_common_fill_super_succeeded) {
1162 /* Only if client_common_fill_super succeeded */
1163 client_common_put_super(sb);
1167 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1168 class_manual_cleanup(obd);
1171 if (sbi->ll_flags & LL_SBI_VERBOSE)
1172 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1175 class_del_profile(profilenm);
1177 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1178 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1179 bdi_destroy(&lsi->lsi_bdi);
1180 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1185 lsi->lsi_llsbi = NULL;
1187 lustre_common_put_super(sb);
1189 cl_env_cache_purge(~0);
1191 module_put(THIS_MODULE);
1194 } /* client_put_super */
1196 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1198 struct inode *inode = NULL;
1200 /* NOTE: we depend on atomic igrab() -bzzz */
1201 lock_res_and_lock(lock);
1202 if (lock->l_resource->lr_lvb_inode) {
1203 struct ll_inode_info * lli;
1204 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1205 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1206 inode = igrab(lock->l_resource->lr_lvb_inode);
1208 inode = lock->l_resource->lr_lvb_inode;
1209 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1210 D_WARNING, lock, "lr_lvb_inode %p is "
1211 "bogus: magic %08x",
1212 lock->l_resource->lr_lvb_inode,
1213 lli->lli_inode_magic);
1217 unlock_res_and_lock(lock);
1221 void ll_dir_clear_lsm_md(struct inode *inode)
1223 struct ll_inode_info *lli = ll_i2info(inode);
1225 LASSERT(S_ISDIR(inode->i_mode));
1227 if (lli->lli_lsm_md != NULL) {
1228 lmv_free_memmd(lli->lli_lsm_md);
1229 lli->lli_lsm_md = NULL;
1233 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1234 const struct lu_fid *fid,
1235 struct lustre_md *md)
1237 struct ll_sb_info *sbi = ll_s2sbi(sb);
1238 struct mdt_body *body = md->body;
1239 struct inode *inode;
1243 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1244 inode = iget_locked(sb, ino);
1245 if (inode == NULL) {
1246 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1247 ll_get_fsname(sb, NULL, 0), PFID(fid));
1248 RETURN(ERR_PTR(-ENOENT));
1251 if (inode->i_state & I_NEW) {
1252 struct ll_inode_info *lli = ll_i2info(inode);
1253 struct lmv_stripe_md *lsm = md->lmv;
1255 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1256 (body->mbo_mode & S_IFMT);
1257 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1260 LTIME_S(inode->i_mtime) = 0;
1261 LTIME_S(inode->i_atime) = 0;
1262 LTIME_S(inode->i_ctime) = 0;
1265 #ifdef HAVE_BACKING_DEV_INFO
1266 /* initializing backing dev info. */
1267 inode->i_mapping->backing_dev_info =
1268 &s2lsi(inode->i_sb)->lsi_bdi;
1270 inode->i_op = &ll_dir_inode_operations;
1271 inode->i_fop = &ll_dir_operations;
1272 lli->lli_fid = *fid;
1275 LASSERT(lsm != NULL);
1276 /* master object FID */
1277 lli->lli_pfid = body->mbo_fid1;
1278 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1279 lli, PFID(fid), PFID(&lli->lli_pfid));
1280 unlock_new_inode(inode);
1286 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1289 struct lmv_stripe_md *lsm = md->lmv;
1292 LASSERT(lsm != NULL);
1293 /* XXX sigh, this lsm_root initialization should be in
1294 * LMV layer, but it needs ll_iget right now, so we
1295 * put this here right now. */
1296 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1297 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1298 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1299 /* Unfortunately ll_iget will call ll_update_inode,
1300 * where the initialization of slave inode is slightly
1301 * different, so it reset lsm_md to NULL to avoid
1302 * initializing lsm for slave inode. */
1303 /* For migrating inode, master stripe and master object will
1304 * be same, so we only need assign this inode */
1305 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1306 lsm->lsm_md_oinfo[i].lmo_root = inode;
1308 lsm->lsm_md_oinfo[i].lmo_root =
1309 ll_iget_anon_dir(inode->i_sb, fid, md);
1311 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1312 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1314 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1322 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1323 const struct lmv_stripe_md *lsm_md2)
1325 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1326 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1327 lsm_md1->lsm_md_master_mdt_index ==
1328 lsm_md2->lsm_md_master_mdt_index &&
1329 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1330 lsm_md1->lsm_md_layout_version ==
1331 lsm_md2->lsm_md_layout_version &&
1332 strcmp(lsm_md1->lsm_md_pool_name,
1333 lsm_md2->lsm_md_pool_name) == 0;
1336 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1338 struct ll_inode_info *lli = ll_i2info(inode);
1339 struct lmv_stripe_md *lsm = md->lmv;
1343 LASSERT(S_ISDIR(inode->i_mode));
1344 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1345 PFID(ll_inode2fid(inode)));
1347 /* no striped information from request. */
1349 if (lli->lli_lsm_md == NULL) {
1351 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1352 LMV_HASH_FLAG_MIGRATION) {
1353 /* migration is done, the temporay MIGRATE layout has
1355 CDEBUG(D_INODE, DFID" finish migration.\n",
1356 PFID(ll_inode2fid(inode)));
1357 lmv_free_memmd(lli->lli_lsm_md);
1358 lli->lli_lsm_md = NULL;
1361 /* The lustre_md from req does not include stripeEA,
1362 * see ll_md_setattr */
1367 /* set the directory layout */
1368 if (lli->lli_lsm_md == NULL) {
1369 struct cl_attr *attr;
1371 rc = ll_init_lsm_md(inode, md);
1375 /* set md->lmv to NULL, so the following free lustre_md
1376 * will not free this lsm */
1378 lli->lli_lsm_md = lsm;
1380 OBD_ALLOC_PTR(attr);
1384 /* validate the lsm */
1385 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1386 ll_md_blocking_ast);
1392 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1393 md->body->mbo_nlink = attr->cat_nlink;
1394 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1395 md->body->mbo_size = attr->cat_size;
1396 if (md->body->mbo_valid & OBD_MD_FLATIME)
1397 md->body->mbo_atime = attr->cat_atime;
1398 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1399 md->body->mbo_ctime = attr->cat_ctime;
1400 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1401 md->body->mbo_mtime = attr->cat_mtime;
1405 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1406 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1410 /* Compare the old and new stripe information */
1411 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1412 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1415 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1416 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1417 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1418 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1419 inode, lsm, old_lsm,
1420 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1421 lsm->lsm_md_stripe_count,
1422 old_lsm->lsm_md_stripe_count,
1423 lsm->lsm_md_master_mdt_index,
1424 old_lsm->lsm_md_master_mdt_index,
1425 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1426 lsm->lsm_md_layout_version,
1427 old_lsm->lsm_md_layout_version,
1428 lsm->lsm_md_pool_name,
1429 old_lsm->lsm_md_pool_name);
1431 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1432 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1433 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1434 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1437 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1438 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1439 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1440 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1449 void ll_clear_inode(struct inode *inode)
1451 struct ll_inode_info *lli = ll_i2info(inode);
1452 struct ll_sb_info *sbi = ll_i2sbi(inode);
1455 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1456 PFID(ll_inode2fid(inode)), inode);
1458 if (S_ISDIR(inode->i_mode)) {
1459 /* these should have been cleared in ll_file_release */
1460 LASSERT(lli->lli_opendir_key == NULL);
1461 LASSERT(lli->lli_sai == NULL);
1462 LASSERT(lli->lli_opendir_pid == 0);
1465 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1467 LASSERT(!lli->lli_open_fd_write_count);
1468 LASSERT(!lli->lli_open_fd_read_count);
1469 LASSERT(!lli->lli_open_fd_exec_count);
1471 if (lli->lli_mds_write_och)
1472 ll_md_real_close(inode, FMODE_WRITE);
1473 if (lli->lli_mds_exec_och)
1474 ll_md_real_close(inode, FMODE_EXEC);
1475 if (lli->lli_mds_read_och)
1476 ll_md_real_close(inode, FMODE_READ);
1478 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1479 OBD_FREE(lli->lli_symlink_name,
1480 strlen(lli->lli_symlink_name) + 1);
1481 lli->lli_symlink_name = NULL;
1484 ll_xattr_cache_destroy(inode);
1486 #ifdef CONFIG_FS_POSIX_ACL
1487 forget_all_cached_acls(inode);
1488 if (lli->lli_posix_acl) {
1489 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1490 posix_acl_release(lli->lli_posix_acl);
1491 lli->lli_posix_acl = NULL;
1494 lli->lli_inode_magic = LLI_INODE_DEAD;
1496 if (S_ISDIR(inode->i_mode))
1497 ll_dir_clear_lsm_md(inode);
1498 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1499 LASSERT(list_empty(&lli->lli_agl_list));
1502 * XXX This has to be done before lsm is freed below, because
1503 * cl_object still uses inode lsm.
1505 cl_inode_fini(inode);
1510 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1512 struct lustre_md md;
1513 struct inode *inode = dentry->d_inode;
1514 struct ll_sb_info *sbi = ll_i2sbi(inode);
1515 struct ptlrpc_request *request = NULL;
1519 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1520 LUSTRE_OPC_ANY, NULL);
1521 if (IS_ERR(op_data))
1522 RETURN(PTR_ERR(op_data));
1524 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1526 ptlrpc_req_finished(request);
1527 if (rc == -ENOENT) {
1529 /* Unlinked special device node? Or just a race?
1530 * Pretend we done everything. */
1531 if (!S_ISREG(inode->i_mode) &&
1532 !S_ISDIR(inode->i_mode)) {
1533 ia_valid = op_data->op_attr.ia_valid;
1534 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1535 rc = simple_setattr(dentry, &op_data->op_attr);
1536 op_data->op_attr.ia_valid = ia_valid;
1538 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1539 CERROR("md_setattr fails: rc = %d\n", rc);
1544 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1545 sbi->ll_md_exp, &md);
1547 ptlrpc_req_finished(request);
1551 ia_valid = op_data->op_attr.ia_valid;
1552 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1553 * cache is not cleared yet. */
1554 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1555 if (S_ISREG(inode->i_mode))
1557 rc = simple_setattr(dentry, &op_data->op_attr);
1558 if (S_ISREG(inode->i_mode))
1559 inode_unlock(inode);
1560 op_data->op_attr.ia_valid = ia_valid;
1562 rc = ll_update_inode(inode, &md);
1563 ptlrpc_req_finished(request);
1568 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1569 * object(s) determine the file size and mtime. Otherwise, the MDS will
1570 * keep these values until such a time that objects are allocated for it.
1571 * We do the MDS operations first, as it is checking permissions for us.
1572 * We don't to the MDS RPC if there is nothing that we want to store there,
1573 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1574 * going to do an RPC anyways.
1576 * If we are doing a truncate, we will send the mtime and ctime updates
1577 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1578 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1581 * In case of HSMimport, we only set attr on MDS.
1583 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1585 struct inode *inode = dentry->d_inode;
1586 struct ll_inode_info *lli = ll_i2info(inode);
1587 struct md_op_data *op_data = NULL;
1591 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1592 "valid %x, hsm_import %d\n",
1593 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1594 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1597 if (attr->ia_valid & ATTR_SIZE) {
1598 /* Check new size against VFS/VM file size limit and rlimit */
1599 rc = inode_newsize_ok(inode, attr->ia_size);
1603 /* The maximum Lustre file size is variable, based on the
1604 * OST maximum object size and number of stripes. This
1605 * needs another check in addition to the VFS check above. */
1606 if (attr->ia_size > ll_file_maxbytes(inode)) {
1607 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1608 PFID(&lli->lli_fid), attr->ia_size,
1609 ll_file_maxbytes(inode));
1613 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1616 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1617 if (attr->ia_valid & TIMES_SET_FLAGS) {
1618 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1619 !cfs_capable(CFS_CAP_FOWNER))
1623 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1624 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1625 (attr->ia_valid & ATTR_CTIME)) {
1626 attr->ia_ctime = current_time(inode);
1627 attr->ia_valid |= ATTR_CTIME_SET;
1629 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1630 (attr->ia_valid & ATTR_ATIME)) {
1631 attr->ia_atime = current_time(inode);
1632 attr->ia_valid |= ATTR_ATIME_SET;
1634 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1635 (attr->ia_valid & ATTR_MTIME)) {
1636 attr->ia_mtime = current_time(inode);
1637 attr->ia_valid |= ATTR_MTIME_SET;
1640 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1641 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1642 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1643 (s64)ktime_get_real_seconds());
1645 if (S_ISREG(inode->i_mode)) {
1646 if (attr->ia_valid & ATTR_SIZE)
1647 inode_dio_write_done(inode);
1648 inode_unlock(inode);
1651 /* We always do an MDS RPC, even if we're only changing the size;
1652 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1654 OBD_ALLOC_PTR(op_data);
1655 if (op_data == NULL)
1656 GOTO(out, rc = -ENOMEM);
1658 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1659 /* If we are changing file size, file content is
1660 * modified, flag it. */
1661 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1662 op_data->op_bias |= MDS_DATA_MODIFIED;
1663 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1666 op_data->op_attr = *attr;
1668 rc = ll_md_setattr(dentry, op_data);
1672 if (!S_ISREG(inode->i_mode) || hsm_import)
1675 if (attr->ia_valid & (ATTR_SIZE |
1676 ATTR_ATIME | ATTR_ATIME_SET |
1677 ATTR_MTIME | ATTR_MTIME_SET |
1678 ATTR_CTIME | ATTR_CTIME_SET)) {
1679 /* For truncate and utimes sending attributes to OSTs, setting
1680 * mtime/atime to the past will be performed under PW [0:EOF]
1681 * extent lock (new_size:EOF for truncate). It may seem
1682 * excessive to send mtime/atime updates to OSTs when not
1683 * setting times to past, but it is necessary due to possible
1684 * time de-synchronization between MDT inode and OST objects */
1685 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1688 /* If the file was restored, it needs to set dirty flag.
1690 * We've already sent MDS_DATA_MODIFIED flag in
1691 * ll_md_setattr() for truncate. However, the MDT refuses to
1692 * set the HS_DIRTY flag on released files, so we have to set
1693 * it again if the file has been restored. Please check how
1694 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1696 * Please notice that if the file is not released, the previous
1697 * MDS_DATA_MODIFIED has taken effect and usually
1698 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1699 * This way we can save an RPC for common open + trunc
1701 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1702 struct hsm_state_set hss = {
1703 .hss_valid = HSS_SETMASK,
1704 .hss_setmask = HS_DIRTY,
1708 rc2 = ll_hsm_state_set(inode, &hss);
1709 /* truncate and write can happen at the same time, so that
1710 * the file can be set modified even though the file is not
1711 * restored from released state, and ll_hsm_state_set() is
1712 * not applicable for the file, and rc2 < 0 is normal in this
1715 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1716 PFID(ll_inode2fid(inode)), rc2);
1721 if (op_data != NULL)
1722 ll_finish_md_op_data(op_data);
1724 if (S_ISREG(inode->i_mode)) {
1726 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1727 inode_dio_wait(inode);
1728 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1729 * flag. ll_update_inode (called from ll_md_setattr), clears
1730 * inode flags, so there is a gap where S_NOSEC is not set.
1731 * This can cause a writer to take the i_mutex unnecessarily,
1732 * but this is safe to do and should be rare. */
1733 inode_has_no_xattr(inode);
1736 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1737 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1742 int ll_setattr(struct dentry *de, struct iattr *attr)
1744 int mode = de->d_inode->i_mode;
1746 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1747 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1748 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1750 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1751 (ATTR_SIZE|ATTR_MODE)) &&
1752 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1753 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1754 !(attr->ia_mode & S_ISGID))))
1755 attr->ia_valid |= ATTR_FORCE;
1757 if ((attr->ia_valid & ATTR_MODE) &&
1759 !(attr->ia_mode & S_ISUID) &&
1760 !(attr->ia_valid & ATTR_KILL_SUID))
1761 attr->ia_valid |= ATTR_KILL_SUID;
1763 if ((attr->ia_valid & ATTR_MODE) &&
1764 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1765 !(attr->ia_mode & S_ISGID) &&
1766 !(attr->ia_valid & ATTR_KILL_SGID))
1767 attr->ia_valid |= ATTR_KILL_SGID;
1769 /* avoid polluted from ATTR_TIMES_SET,
1770 * projid is not expected to be set here */
1771 attr->ia_valid &= ~MDS_ATTR_PROJID;
1773 return ll_setattr_raw(de, attr, false);
1776 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1779 struct ll_sb_info *sbi = ll_s2sbi(sb);
1780 struct obd_statfs obd_osfs;
1785 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1787 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1789 CERROR("md_statfs fails: rc = %d\n", rc);
1793 osfs->os_type = sb->s_magic;
1795 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1796 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1798 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1799 flags |= OBD_STATFS_NODELAY;
1801 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1803 CERROR("obd_statfs fails: rc = %d\n", rc);
1807 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1808 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1811 osfs->os_bsize = obd_osfs.os_bsize;
1812 osfs->os_blocks = obd_osfs.os_blocks;
1813 osfs->os_bfree = obd_osfs.os_bfree;
1814 osfs->os_bavail = obd_osfs.os_bavail;
1816 /* If we don't have as many objects free on the OST as inodes
1817 * on the MDS, we reduce the total number of inodes to
1818 * compensate, so that the "inodes in use" number is correct.
1820 if (obd_osfs.os_ffree < osfs->os_ffree) {
1821 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1823 osfs->os_ffree = obd_osfs.os_ffree;
1828 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1830 struct super_block *sb = de->d_sb;
1831 struct obd_statfs osfs;
1832 __u64 fsid = huge_encode_dev(sb->s_dev);
1835 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1836 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1838 /* Some amount of caching on the client is allowed */
1839 rc = ll_statfs_internal(sb, &osfs, 0);
1843 statfs_unpack(sfs, &osfs);
1845 /* We need to downshift for all 32-bit kernels, because we can't
1846 * tell if the kernel is being called via sys_statfs64() or not.
1847 * Stop before overflowing f_bsize - in which case it is better
1848 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1849 if (sizeof(long) < 8) {
1850 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1853 osfs.os_blocks >>= 1;
1854 osfs.os_bfree >>= 1;
1855 osfs.os_bavail >>= 1;
1859 sfs->f_blocks = osfs.os_blocks;
1860 sfs->f_bfree = osfs.os_bfree;
1861 sfs->f_bavail = osfs.os_bavail;
1862 sfs->f_fsid.val[0] = (__u32)fsid;
1863 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1867 void ll_inode_size_lock(struct inode *inode)
1869 struct ll_inode_info *lli;
1871 LASSERT(!S_ISDIR(inode->i_mode));
1873 lli = ll_i2info(inode);
1874 mutex_lock(&lli->lli_size_mutex);
1877 void ll_inode_size_unlock(struct inode *inode)
1879 struct ll_inode_info *lli;
1881 lli = ll_i2info(inode);
1882 mutex_unlock(&lli->lli_size_mutex);
1885 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1887 struct ll_inode_info *lli = ll_i2info(inode);
1888 struct mdt_body *body = md->body;
1889 struct ll_sb_info *sbi = ll_i2sbi(inode);
1892 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1893 rc = cl_file_inode_init(inode, md);
1898 if (S_ISDIR(inode->i_mode)) {
1899 rc = ll_update_lsm_md(inode, md);
1904 #ifdef CONFIG_FS_POSIX_ACL
1905 if (body->mbo_valid & OBD_MD_FLACL) {
1906 spin_lock(&lli->lli_lock);
1907 if (lli->lli_posix_acl)
1908 posix_acl_release(lli->lli_posix_acl);
1909 lli->lli_posix_acl = md->posix_acl;
1910 spin_unlock(&lli->lli_lock);
1913 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1914 sbi->ll_flags & LL_SBI_32BIT_API);
1915 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1917 if (body->mbo_valid & OBD_MD_FLATIME) {
1918 if (body->mbo_atime > LTIME_S(inode->i_atime))
1919 LTIME_S(inode->i_atime) = body->mbo_atime;
1920 lli->lli_atime = body->mbo_atime;
1923 if (body->mbo_valid & OBD_MD_FLMTIME) {
1924 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1925 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1926 "to %llu\n", inode->i_ino,
1927 LTIME_S(inode->i_mtime), body->mbo_mtime);
1928 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1930 lli->lli_mtime = body->mbo_mtime;
1933 if (body->mbo_valid & OBD_MD_FLCTIME) {
1934 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1935 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1936 lli->lli_ctime = body->mbo_ctime;
1939 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1940 if (body->mbo_valid & OBD_MD_FLFLAGS)
1941 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1942 if (body->mbo_valid & OBD_MD_FLMODE)
1943 inode->i_mode = (inode->i_mode & S_IFMT) |
1944 (body->mbo_mode & ~S_IFMT);
1946 if (body->mbo_valid & OBD_MD_FLTYPE)
1947 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1948 (body->mbo_mode & S_IFMT);
1950 LASSERT(inode->i_mode != 0);
1951 if (S_ISREG(inode->i_mode))
1952 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1953 LL_MAX_BLKSIZE_BITS);
1955 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1957 if (body->mbo_valid & OBD_MD_FLUID)
1958 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1959 if (body->mbo_valid & OBD_MD_FLGID)
1960 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1961 if (body->mbo_valid & OBD_MD_FLPROJID)
1962 lli->lli_projid = body->mbo_projid;
1963 if (body->mbo_valid & OBD_MD_FLNLINK)
1964 set_nlink(inode, body->mbo_nlink);
1965 if (body->mbo_valid & OBD_MD_FLRDEV)
1966 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1968 if (body->mbo_valid & OBD_MD_FLID) {
1969 /* FID shouldn't be changed! */
1970 if (fid_is_sane(&lli->lli_fid)) {
1971 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1972 "Trying to change FID "DFID
1973 " to the "DFID", inode "DFID"(%p)\n",
1974 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1975 PFID(ll_inode2fid(inode)), inode);
1977 lli->lli_fid = body->mbo_fid1;
1981 LASSERT(fid_seq(&lli->lli_fid) != 0);
1983 if (body->mbo_valid & OBD_MD_FLSIZE) {
1984 i_size_write(inode, body->mbo_size);
1986 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1987 PFID(ll_inode2fid(inode)),
1988 (unsigned long long)body->mbo_size);
1990 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1991 inode->i_blocks = body->mbo_blocks;
1994 if (body->mbo_valid & OBD_MD_TSTATE) {
1995 /* Set LLIF_FILE_RESTORING if restore ongoing and
1996 * clear it when done to ensure to start again
1997 * glimpsing updated attrs
1999 if (body->mbo_t_state & MS_RESTORE)
2000 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2002 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2008 int ll_read_inode2(struct inode *inode, void *opaque)
2010 struct lustre_md *md = opaque;
2011 struct ll_inode_info *lli = ll_i2info(inode);
2015 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2016 PFID(&lli->lli_fid), inode);
2018 /* Core attributes from the MDS first. This is a new inode, and
2019 * the VFS doesn't zero times in the core inode so we have to do
2020 * it ourselves. They will be overwritten by either MDS or OST
2021 * attributes - we just need to make sure they aren't newer. */
2022 LTIME_S(inode->i_mtime) = 0;
2023 LTIME_S(inode->i_atime) = 0;
2024 LTIME_S(inode->i_ctime) = 0;
2026 rc = ll_update_inode(inode, md);
2030 /* OIDEBUG(inode); */
2032 #ifdef HAVE_BACKING_DEV_INFO
2033 /* initializing backing dev info. */
2034 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2036 if (S_ISREG(inode->i_mode)) {
2037 struct ll_sb_info *sbi = ll_i2sbi(inode);
2038 inode->i_op = &ll_file_inode_operations;
2039 inode->i_fop = sbi->ll_fop;
2040 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2042 } else if (S_ISDIR(inode->i_mode)) {
2043 inode->i_op = &ll_dir_inode_operations;
2044 inode->i_fop = &ll_dir_operations;
2046 } else if (S_ISLNK(inode->i_mode)) {
2047 inode->i_op = &ll_fast_symlink_inode_operations;
2050 inode->i_op = &ll_special_inode_operations;
2052 init_special_inode(inode, inode->i_mode,
2061 void ll_delete_inode(struct inode *inode)
2063 struct ll_inode_info *lli = ll_i2info(inode);
2066 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2067 /* It is last chance to write out dirty pages,
2068 * otherwise we may lose data while umount */
2069 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2071 truncate_inode_pages_final(&inode->i_data);
2073 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2074 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2075 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2077 #ifdef HAVE_SBOPS_EVICT_INODE
2078 ll_clear_inode(inode);
2085 int ll_iocontrol(struct inode *inode, struct file *file,
2086 unsigned int cmd, unsigned long arg)
2088 struct ll_sb_info *sbi = ll_i2sbi(inode);
2089 struct ptlrpc_request *req = NULL;
2094 case FS_IOC_GETFLAGS: {
2095 struct mdt_body *body;
2096 struct md_op_data *op_data;
2098 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2099 0, 0, LUSTRE_OPC_ANY,
2101 if (IS_ERR(op_data))
2102 RETURN(PTR_ERR(op_data));
2104 op_data->op_valid = OBD_MD_FLFLAGS;
2105 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2106 ll_finish_md_op_data(op_data);
2108 CERROR("%s: failure inode "DFID": rc = %d\n",
2109 sbi->ll_md_exp->exp_obd->obd_name,
2110 PFID(ll_inode2fid(inode)), rc);
2114 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2116 flags = body->mbo_flags;
2118 ptlrpc_req_finished(req);
2120 RETURN(put_user(flags, (int __user *)arg));
2122 case FS_IOC_SETFLAGS: {
2124 struct md_op_data *op_data;
2125 struct cl_object *obj;
2127 if (get_user(flags, (int __user *)arg))
2130 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2131 LUSTRE_OPC_ANY, NULL);
2132 if (IS_ERR(op_data))
2133 RETURN(PTR_ERR(op_data));
2135 op_data->op_attr_flags = flags;
2136 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2137 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2138 ll_finish_md_op_data(op_data);
2139 ptlrpc_req_finished(req);
2143 inode->i_flags = ll_ext_to_inode_flags(flags);
2145 obj = ll_i2info(inode)->lli_clob;
2149 OBD_ALLOC_PTR(attr);
2153 attr->ia_valid = ATTR_ATTR_FLAG;
2154 rc = cl_setattr_ost(obj, attr, flags);
2166 int ll_flush_ctx(struct inode *inode)
2168 struct ll_sb_info *sbi = ll_i2sbi(inode);
2170 CDEBUG(D_SEC, "flush context for user %d\n",
2171 from_kuid(&init_user_ns, current_uid()));
2173 obd_set_info_async(NULL, sbi->ll_md_exp,
2174 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2176 obd_set_info_async(NULL, sbi->ll_dt_exp,
2177 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2182 /* umount -f client means force down, don't save state */
2183 void ll_umount_begin(struct super_block *sb)
2185 struct ll_sb_info *sbi = ll_s2sbi(sb);
2186 struct obd_device *obd;
2187 struct obd_ioctl_data *ioc_data;
2188 struct l_wait_info lwi;
2189 wait_queue_head_t waitq;
2192 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2193 sb->s_count, atomic_read(&sb->s_active));
2195 obd = class_exp2obd(sbi->ll_md_exp);
2197 CERROR("Invalid MDC connection handle %#llx\n",
2198 sbi->ll_md_exp->exp_handle.h_cookie);
2204 obd = class_exp2obd(sbi->ll_dt_exp);
2206 CERROR("Invalid LOV connection handle %#llx\n",
2207 sbi->ll_dt_exp->exp_handle.h_cookie);
2213 OBD_ALLOC_PTR(ioc_data);
2215 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2216 sizeof *ioc_data, ioc_data, NULL);
2218 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2219 sizeof *ioc_data, ioc_data, NULL);
2221 OBD_FREE_PTR(ioc_data);
2224 /* Really, we'd like to wait until there are no requests outstanding,
2225 * and then continue. For now, we just periodically checking for vfs
2226 * to decrement mnt_cnt and hope to finish it within 10sec.
2228 init_waitqueue_head(&waitq);
2229 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2230 cfs_time_seconds(1), NULL, NULL);
2231 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2236 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2238 struct ll_sb_info *sbi = ll_s2sbi(sb);
2239 char *profilenm = get_profile_name(sb);
2243 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2244 read_only = *flags & MS_RDONLY;
2245 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2246 sizeof(KEY_READ_ONLY),
2247 KEY_READ_ONLY, sizeof(read_only),
2250 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2251 profilenm, read_only ?
2252 "read-only" : "read-write", err);
2257 sb->s_flags |= MS_RDONLY;
2259 sb->s_flags &= ~MS_RDONLY;
2261 if (sbi->ll_flags & LL_SBI_VERBOSE)
2262 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2263 read_only ? "read-only" : "read-write");
2269 * Cleanup the open handle that is cached on MDT-side.
2271 * For open case, the client side open handling thread may hit error
2272 * after the MDT grant the open. Under such case, the client should
2273 * send close RPC to the MDT as cleanup; otherwise, the open handle
2274 * on the MDT will be leaked there until the client umount or evicted.
2276 * In further, if someone unlinked the file, because the open handle
2277 * holds the reference on such file/object, then it will block the
2278 * subsequent threads that want to locate such object via FID.
2280 * \param[in] sb super block for this file-system
2281 * \param[in] open_req pointer to the original open request
2283 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2285 struct mdt_body *body;
2286 struct md_op_data *op_data;
2287 struct ptlrpc_request *close_req = NULL;
2288 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2291 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2292 OBD_ALLOC_PTR(op_data);
2293 if (op_data == NULL) {
2294 CWARN("%s: cannot allocate op_data to release open handle for "
2296 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2301 op_data->op_fid1 = body->mbo_fid1;
2302 op_data->op_handle = body->mbo_handle;
2303 op_data->op_mod_time = ktime_get_real_seconds();
2304 md_close(exp, op_data, NULL, &close_req);
2305 ptlrpc_req_finished(close_req);
2306 ll_finish_md_op_data(op_data);
2311 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2312 struct super_block *sb, struct lookup_intent *it)
2314 struct ll_sb_info *sbi = NULL;
2315 struct lustre_md md = { NULL };
2319 LASSERT(*inode || sb);
2320 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2321 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2322 sbi->ll_md_exp, &md);
2327 rc = ll_update_inode(*inode, &md);
2331 LASSERT(sb != NULL);
2334 * At this point server returns to client's same fid as client
2335 * generated for creating. So using ->fid1 is okay here.
2337 if (!fid_is_sane(&md.body->mbo_fid1)) {
2338 CERROR("%s: Fid is insane "DFID"\n",
2339 ll_get_fsname(sb, NULL, 0),
2340 PFID(&md.body->mbo_fid1));
2341 GOTO(out, rc = -EINVAL);
2344 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2345 sbi->ll_flags & LL_SBI_32BIT_API),
2347 if (IS_ERR(*inode)) {
2348 #ifdef CONFIG_FS_POSIX_ACL
2350 posix_acl_release(md.posix_acl);
2351 md.posix_acl = NULL;
2354 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2356 CERROR("new_inode -fatal: rc %d\n", rc);
2361 /* Handling piggyback layout lock.
2362 * Layout lock can be piggybacked by getattr and open request.
2363 * The lsm can be applied to inode only if it comes with a layout lock
2364 * otherwise correct layout may be overwritten, for example:
2365 * 1. proc1: mdt returns a lsm but not granting layout
2366 * 2. layout was changed by another client
2367 * 3. proc2: refresh layout and layout lock granted
2368 * 4. proc1: to apply a stale layout */
2369 if (it != NULL && it->it_lock_mode != 0) {
2370 struct lustre_handle lockh;
2371 struct ldlm_lock *lock;
2373 lockh.cookie = it->it_lock_handle;
2374 lock = ldlm_handle2lock(&lockh);
2375 LASSERT(lock != NULL);
2376 if (ldlm_has_layout(lock)) {
2377 struct cl_object_conf conf;
2379 memset(&conf, 0, sizeof(conf));
2380 conf.coc_opc = OBJECT_CONF_SET;
2381 conf.coc_inode = *inode;
2382 conf.coc_lock = lock;
2383 conf.u.coc_layout = md.layout;
2384 (void)ll_layout_conf(*inode, &conf);
2386 LDLM_LOCK_PUT(lock);
2392 md_free_lustre_md(sbi->ll_md_exp, &md);
2395 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2396 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2401 int ll_obd_statfs(struct inode *inode, void __user *arg)
2403 struct ll_sb_info *sbi = NULL;
2404 struct obd_export *exp;
2406 struct obd_ioctl_data *data = NULL;
2410 if (!inode || !(sbi = ll_i2sbi(inode)))
2411 GOTO(out_statfs, rc = -EINVAL);
2413 rc = obd_ioctl_getdata(&buf, &len, arg);
2415 GOTO(out_statfs, rc);
2418 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2419 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2420 GOTO(out_statfs, rc = -EINVAL);
2422 if (data->ioc_inllen1 != sizeof(__u32) ||
2423 data->ioc_inllen2 != sizeof(__u32) ||
2424 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2425 data->ioc_plen2 != sizeof(struct obd_uuid))
2426 GOTO(out_statfs, rc = -EINVAL);
2428 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2429 if (type & LL_STATFS_LMV)
2430 exp = sbi->ll_md_exp;
2431 else if (type & LL_STATFS_LOV)
2432 exp = sbi->ll_dt_exp;
2434 GOTO(out_statfs, rc = -ENODEV);
2436 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2438 GOTO(out_statfs, rc);
2440 OBD_FREE_LARGE(buf, len);
2444 int ll_process_config(struct lustre_cfg *lcfg)
2446 struct super_block *sb;
2451 /* The instance name contains the sb: lustre-client-aacfe000 */
2452 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2453 if (!ptr || !*(++ptr))
2455 if (sscanf(ptr, "%lx", &x) != 1)
2457 sb = (struct super_block *)x;
2458 /* This better be a real Lustre superblock! */
2459 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2461 /* Note we have not called client_common_fill_super yet, so
2462 proc fns must be able to handle that! */
2463 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2470 /* this function prepares md_op_data hint for passing it down to MD stack. */
2471 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2472 struct inode *i1, struct inode *i2,
2473 const char *name, size_t namelen,
2474 __u32 mode, __u32 opc, void *data)
2476 LASSERT(i1 != NULL);
2479 /* Do not reuse namelen for something else. */
2481 return ERR_PTR(-EINVAL);
2483 if (namelen > ll_i2sbi(i1)->ll_namelen)
2484 return ERR_PTR(-ENAMETOOLONG);
2486 if (!lu_name_is_valid_2(name, namelen))
2487 return ERR_PTR(-EINVAL);
2490 if (op_data == NULL)
2491 OBD_ALLOC_PTR(op_data);
2493 if (op_data == NULL)
2494 return ERR_PTR(-ENOMEM);
2496 ll_i2gids(op_data->op_suppgids, i1, i2);
2497 op_data->op_fid1 = *ll_inode2fid(i1);
2498 op_data->op_default_stripe_offset = -1;
2499 if (S_ISDIR(i1->i_mode)) {
2500 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2501 if (opc == LUSTRE_OPC_MKDIR)
2502 op_data->op_default_stripe_offset =
2503 ll_i2info(i1)->lli_def_stripe_offset;
2507 op_data->op_fid2 = *ll_inode2fid(i2);
2508 if (S_ISDIR(i2->i_mode))
2509 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2511 fid_zero(&op_data->op_fid2);
2514 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2515 op_data->op_cli_flags |= CLI_HASH64;
2517 if (ll_need_32bit_api(ll_i2sbi(i1)))
2518 op_data->op_cli_flags |= CLI_API32;
2520 op_data->op_name = name;
2521 op_data->op_namelen = namelen;
2522 op_data->op_mode = mode;
2523 op_data->op_mod_time = ktime_get_real_seconds();
2524 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2525 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2526 op_data->op_cap = cfs_curproc_cap_pack();
2527 op_data->op_mds = 0;
2528 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2529 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2530 op_data->op_bias |= MDS_CREATE_VOLATILE;
2532 op_data->op_data = data;
2537 void ll_finish_md_op_data(struct md_op_data *op_data)
2539 security_release_secctx(op_data->op_file_secctx,
2540 op_data->op_file_secctx_size);
2541 OBD_FREE_PTR(op_data);
2544 #ifdef HAVE_SUPEROPS_USE_DENTRY
2545 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2547 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2550 struct ll_sb_info *sbi;
2552 #ifdef HAVE_SUPEROPS_USE_DENTRY
2553 LASSERT((seq != NULL) && (dentry != NULL));
2554 sbi = ll_s2sbi(dentry->d_sb);
2556 LASSERT((seq != NULL) && (vfs != NULL));
2557 sbi = ll_s2sbi(vfs->mnt_sb);
2560 if (sbi->ll_flags & LL_SBI_NOLCK)
2561 seq_puts(seq, ",nolock");
2563 if (sbi->ll_flags & LL_SBI_FLOCK)
2564 seq_puts(seq, ",flock");
2566 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2567 seq_puts(seq, ",localflock");
2569 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2570 seq_puts(seq, ",user_xattr");
2572 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2573 seq_puts(seq, ",lazystatfs");
2575 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2576 seq_puts(seq, ",user_fid2path");
2578 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2579 seq_puts(seq, ",always_ping");
2585 * Get obd name by cmd, and copy out to user space
2587 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2589 struct ll_sb_info *sbi = ll_i2sbi(inode);
2590 struct obd_device *obd;
2593 if (cmd == OBD_IOC_GETDTNAME)
2594 obd = class_exp2obd(sbi->ll_dt_exp);
2595 else if (cmd == OBD_IOC_GETMDNAME)
2596 obd = class_exp2obd(sbi->ll_md_exp);
2603 if (copy_to_user((void __user *)arg, obd->obd_name,
2604 strlen(obd->obd_name) + 1))
2611 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2612 * fsname will be returned in this buffer; otherwise, a static buffer will be
2613 * used to store the fsname and returned to caller.
2615 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2617 static char fsname_static[MTI_NAME_MAXLEN];
2618 struct lustre_sb_info *lsi = s2lsi(sb);
2623 /* this means the caller wants to use static buffer
2624 * and it doesn't care about race. Usually this is
2625 * in error reporting path */
2626 buf = fsname_static;
2627 buflen = sizeof(fsname_static);
2630 len = strlen(lsi->lsi_lmd->lmd_profile);
2631 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2632 if (ptr && (strcmp(ptr, "-client") == 0))
2635 if (unlikely(len >= buflen))
2637 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2643 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2650 p.mnt = current->fs->root.mnt;
2652 path = d_path(&p, buf, bufsize);
2657 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2659 char *buf, *path = NULL;
2660 struct dentry *dentry = NULL;
2661 struct inode *inode = page->mapping->host;
2663 /* this can be called inside spin lock so use GFP_ATOMIC. */
2664 buf = (char *)__get_free_page(GFP_ATOMIC);
2666 dentry = d_find_alias(page->mapping->host);
2668 path = ll_d_path(dentry, buf, PAGE_SIZE);
2672 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2673 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2674 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2675 PFID(ll_inode2fid(inode)),
2676 (path && !IS_ERR(path)) ? path : "", ioret);
2682 free_page((unsigned long)buf);
2685 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2686 struct lov_user_md **kbuf)
2688 struct lov_user_md lum;
2692 if (copy_from_user(&lum, md, sizeof(lum)))
2695 lum_size = ll_lov_user_md_size(&lum);
2699 OBD_ALLOC(*kbuf, lum_size);
2703 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2704 OBD_FREE(*kbuf, lum_size);
2712 * Compute llite root squash state after a change of root squash
2713 * configuration setting or add/remove of a lnet nid
2715 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2717 struct root_squash_info *squash = &sbi->ll_squash;
2720 struct lnet_process_id id;
2722 /* Update norootsquash flag */
2723 down_write(&squash->rsi_sem);
2724 if (list_empty(&squash->rsi_nosquash_nids))
2725 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2727 /* Do not apply root squash as soon as one of our NIDs is
2728 * in the nosquash_nids list */
2731 while (LNetGetId(i++, &id) != -ENOENT) {
2732 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2734 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2740 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2742 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2744 up_write(&squash->rsi_sem);
2748 * Parse linkea content to extract information about a given hardlink
2750 * \param[in] ldata - Initialized linkea data
2751 * \param[in] linkno - Link identifier
2752 * \param[out] parent_fid - The entry's parent FID
2753 * \param[out] ln - Entry name destination buffer
2755 * \retval 0 on success
2756 * \retval Appropriate negative error code on failure
2758 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2759 struct lu_fid *parent_fid, struct lu_name *ln)
2765 rc = linkea_init_with_rec(ldata);
2769 if (linkno >= ldata->ld_leh->leh_reccount)
2770 /* beyond last link */
2773 linkea_first_entry(ldata);
2774 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2775 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2780 linkea_next_entry(ldata);
2790 * Get parent FID and name of an identified link. Operation is performed for
2791 * a given link number, letting the caller iterate over linkno to list one or
2792 * all links of an entry.
2794 * \param[in] file - File descriptor against which to perform the operation
2795 * \param[in,out] arg - User-filled structure containing the linkno to operate
2796 * on and the available size. It is eventually filled with
2797 * the requested information or left untouched on error
2799 * \retval - 0 on success
2800 * \retval - Appropriate negative error code on failure
2802 int ll_getparent(struct file *file, struct getparent __user *arg)
2804 struct inode *inode = file_inode(file);
2805 struct linkea_data *ldata;
2806 struct lu_buf buf = LU_BUF_NULL;
2808 struct lu_fid parent_fid;
2815 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2816 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2819 if (get_user(name_size, &arg->gp_name_size))
2822 if (get_user(linkno, &arg->gp_linkno))
2825 if (name_size > PATH_MAX)
2828 OBD_ALLOC(ldata, sizeof(*ldata));
2832 rc = linkea_data_new(ldata, &buf);
2834 GOTO(ldata_free, rc);
2836 #ifdef HAVE_XATTR_HANDLER_FLAGS
2837 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2838 buf.lb_len, OBD_MD_FLXATTR);
2840 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2842 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2846 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2850 if (ln.ln_namelen >= name_size)
2851 GOTO(lb_free, rc = -EOVERFLOW);
2853 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2854 GOTO(lb_free, rc = -EFAULT);
2856 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2857 GOTO(lb_free, rc = -EFAULT);
2859 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2860 GOTO(lb_free, rc = -EFAULT);
2865 OBD_FREE(ldata, sizeof(*ldata));