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_running_max = LL_SA_RUNNING_DEF;
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
134 sbi->ll_flags |= LL_SBI_TINY_WRITE;
137 sbi->ll_squash.rsi_uid = 0;
138 sbi->ll_squash.rsi_gid = 0;
139 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140 init_rwsem(&sbi->ll_squash.rsi_sem);
145 static void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
151 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
152 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
153 if (sbi->ll_cache != NULL) {
154 cl_cache_decref(sbi->ll_cache);
155 sbi->ll_cache = NULL;
157 OBD_FREE(sbi, sizeof(*sbi));
162 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
164 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
165 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169 struct vfsmount *mnt)
171 struct inode *root = NULL;
172 struct ll_sb_info *sbi = ll_s2sbi(sb);
173 struct obd_device *obd;
174 struct obd_statfs *osfs = NULL;
175 struct ptlrpc_request *request = NULL;
176 struct obd_connect_data *data = NULL;
177 struct obd_uuid *uuid;
178 struct md_op_data *op_data;
179 struct lustre_md lmd;
181 int size, err, checksum;
184 obd = class_name2obd(md);
186 CERROR("MD %s: not setup or attached\n", md);
200 /* pass client page size via ocd_grant_blkbits, the server should report
201 * back its backend blocksize for grant calculation purpose */
202 data->ocd_grant_blkbits = PAGE_SHIFT;
204 /* indicate MDT features supported by this client */
205 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
206 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
207 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
208 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
209 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
210 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
211 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
212 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
213 OBD_CONNECT_64BITHASH |
214 OBD_CONNECT_EINPROGRESS |
215 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
216 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
217 OBD_CONNECT_MAX_EASIZE |
218 OBD_CONNECT_FLOCK_DEAD |
219 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
220 OBD_CONNECT_OPEN_BY_FID |
221 OBD_CONNECT_DIR_STRIPE |
222 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
223 OBD_CONNECT_SUBTREE |
224 OBD_CONNECT_MULTIMODRPCS |
225 OBD_CONNECT_GRANT_PARAM |
226 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
228 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
229 OBD_CONNECT2_LOCK_CONVERT |
230 OBD_CONNECT2_DIR_MIGRATE |
231 OBD_CONNECT2_SUM_STATFS;
233 #ifdef HAVE_LRU_RESIZE_SUPPORT
234 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
235 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
237 #ifdef CONFIG_FS_POSIX_ACL
238 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
239 OBD_CONNECT_LARGE_ACL;
242 data->ocd_cksum_types = obd_cksum_types_supported_client();
244 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
245 /* flag mdc connection as lightweight, only used for test
246 * purpose, use with care */
247 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
249 data->ocd_ibits_known = MDS_INODELOCK_FULL;
250 data->ocd_version = LUSTRE_VERSION_CODE;
252 if (sb->s_flags & MS_RDONLY)
253 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
254 if (sbi->ll_flags & LL_SBI_USER_XATTR)
255 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
258 /* Setting this indicates we correctly support S_NOSEC (See kernel
259 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
261 sb->s_flags |= MS_NOSEC;
264 if (sbi->ll_flags & LL_SBI_FLOCK)
265 sbi->ll_fop = &ll_file_operations_flock;
266 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
267 sbi->ll_fop = &ll_file_operations;
269 sbi->ll_fop = &ll_file_operations_noflock;
271 /* always ping even if server suppress_pings */
272 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
273 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
275 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
276 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
277 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
279 data->ocd_brw_size = MD_MAX_BRW_SIZE;
281 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
283 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
284 "recovery, of which this client is not a "
285 "part. Please wait for recovery to complete,"
286 " abort, or time out.\n", md);
289 CERROR("cannot connect to %s: rc = %d\n", md, err);
293 sbi->ll_md_exp->exp_connect_data = *data;
295 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
296 LUSTRE_SEQ_METADATA);
298 CERROR("%s: Can't init metadata layer FID infrastructure, "
299 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
303 /* For mount, we only need fs info from MDT0, and also in DNE, it
304 * can make sure the client can be mounted as long as MDT0 is
306 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
307 ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
308 OBD_STATFS_FOR_MDT0);
310 GOTO(out_md_fid, err);
312 /* This needs to be after statfs to ensure connect has finished.
313 * Note that "data" does NOT contain the valid connect reply.
314 * If connecting to a 1.8 server there will be no LMV device, so
315 * we can access the MDC export directly and exp_connect_flags will
316 * be non-zero, but if accessing an upgraded 2.1 server it will
317 * have the correct flags filled in.
318 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
319 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
320 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
321 valid != CLIENT_CONNECT_MDT_REQD) {
324 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
325 obd_connect_flags2str(buf, PAGE_SIZE,
326 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
327 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
328 "feature(s) needed for correct operation "
329 "of this client (%s). Please upgrade "
330 "server or downgrade client.\n",
331 sbi->ll_md_exp->exp_obd->obd_name, buf);
332 OBD_FREE(buf, PAGE_SIZE);
333 GOTO(out_md_fid, err = -EPROTO);
336 size = sizeof(*data);
337 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
338 KEY_CONN_DATA, &size, data);
340 CERROR("%s: Get connect data failed: rc = %d\n",
341 sbi->ll_md_exp->exp_obd->obd_name, err);
342 GOTO(out_md_fid, err);
345 LASSERT(osfs->os_bsize);
346 sb->s_blocksize = osfs->os_bsize;
347 sb->s_blocksize_bits = log2(osfs->os_bsize);
348 sb->s_magic = LL_SUPER_MAGIC;
349 sb->s_maxbytes = MAX_LFS_FILESIZE;
350 sbi->ll_namelen = osfs->os_namelen;
351 sbi->ll_mnt.mnt = current->fs->root.mnt;
353 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
354 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
355 LCONSOLE_INFO("Disabling user_xattr feature because "
356 "it is not supported on the server\n");
357 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
360 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
362 sb->s_flags |= MS_POSIXACL;
364 sbi->ll_flags |= LL_SBI_ACL;
366 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
368 sb->s_flags &= ~MS_POSIXACL;
370 sbi->ll_flags &= ~LL_SBI_ACL;
373 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
374 sbi->ll_flags |= LL_SBI_64BIT_HASH;
376 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
377 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
379 if (obd_connect_has_secctx(data))
380 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
382 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
383 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
384 LCONSOLE_INFO("%s: disabling xattr cache due to "
385 "unknown maximum xattr size.\n", dt);
386 } else if (!sbi->ll_xattr_cache_set) {
387 /* If xattr_cache is already set (no matter 0 or 1)
388 * during processing llog, it won't be enabled here. */
389 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
390 sbi->ll_xattr_cache_enabled = 1;
394 obd = class_name2obd(dt);
396 CERROR("DT %s: not setup or attached\n", dt);
397 GOTO(out_md_fid, err = -ENODEV);
400 /* pass client page size via ocd_grant_blkbits, the server should report
401 * back its backend blocksize for grant calculation purpose */
402 data->ocd_grant_blkbits = PAGE_SHIFT;
404 /* indicate OST features supported by this client */
405 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
406 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
407 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
408 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
409 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
410 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
411 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
412 OBD_CONNECT_EINPROGRESS |
413 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
414 OBD_CONNECT_LAYOUTLOCK |
415 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
416 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
417 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
419 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
420 * can interoperate with an older version of lockahead which was released prior
421 * to landing in master. This support will be dropped when 2.13 development
422 * starts. At the point, we should not just drop the connect flag (below), we
423 * should also remove the support in the code.
425 * Removing it means a few things:
426 * 1. Remove this section here
427 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
428 * 3. Remove function exp_connect_lockahead_old
429 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
431 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
432 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
435 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
437 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
438 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
440 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
441 * disabled by default, because it can still be enabled on the
442 * fly via /proc. As a consequence, we still need to come to an
443 * agreement on the supported algorithms at connect time */
444 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
446 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
447 data->ocd_cksum_types = OBD_CKSUM_ADLER;
449 data->ocd_cksum_types = obd_cksum_types_supported_client();
451 #ifdef HAVE_LRU_RESIZE_SUPPORT
452 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
454 /* always ping even if server suppress_pings */
455 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
456 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
458 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
459 "ocd_grant: %d\n", data->ocd_connect_flags,
460 data->ocd_version, data->ocd_grant);
462 obd->obd_upcall.onu_owner = &sbi->ll_lco;
463 obd->obd_upcall.onu_upcall = cl_ocd_update;
465 data->ocd_brw_size = DT_MAX_BRW_SIZE;
467 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
470 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
471 "recovery, of which this client is not a "
472 "part. Please wait for recovery to "
473 "complete, abort, or time out.\n", dt);
476 CERROR("%s: Cannot connect to %s: rc = %d\n",
477 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
481 sbi->ll_dt_exp->exp_connect_data = *data;
483 /* Don't change value if it was specified in the config log */
484 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
485 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
486 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
487 (data->ocd_brw_size >> PAGE_SHIFT));
489 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
490 LUSTRE_SEQ_METADATA);
492 CERROR("%s: Can't init data layer FID infrastructure, "
493 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
497 mutex_lock(&sbi->ll_lco.lco_lock);
498 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
499 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
500 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
501 mutex_unlock(&sbi->ll_lco.lco_lock);
503 fid_zero(&sbi->ll_root_fid);
504 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
507 CERROR("cannot mds_connect: rc = %d\n", err);
508 GOTO(out_lock_cn_cb, err);
510 if (!fid_is_sane(&sbi->ll_root_fid)) {
511 CERROR("%s: Invalid root fid "DFID" during mount\n",
512 sbi->ll_md_exp->exp_obd->obd_name,
513 PFID(&sbi->ll_root_fid));
514 GOTO(out_lock_cn_cb, err = -EINVAL);
516 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
518 sb->s_op = &lustre_super_operations;
519 #ifdef HAVE_XATTR_HANDLER_FLAGS
520 sb->s_xattr = ll_xattr_handlers;
522 #if THREAD_SIZE >= 8192 /*b=17630*/
523 sb->s_export_op = &lustre_export_operations;
527 * XXX: move this to after cbd setup? */
528 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
529 if (sbi->ll_flags & LL_SBI_ACL)
530 valid |= OBD_MD_FLACL;
532 OBD_ALLOC_PTR(op_data);
534 GOTO(out_lock_cn_cb, err = -ENOMEM);
536 op_data->op_fid1 = sbi->ll_root_fid;
537 op_data->op_mode = 0;
538 op_data->op_valid = valid;
540 err = md_getattr(sbi->ll_md_exp, op_data, &request);
542 OBD_FREE_PTR(op_data);
544 CERROR("%s: md_getattr failed for root: rc = %d\n",
545 sbi->ll_md_exp->exp_obd->obd_name, err);
546 GOTO(out_lock_cn_cb, err);
549 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
550 sbi->ll_md_exp, &lmd);
552 CERROR("failed to understand root inode md: rc = %d\n", err);
553 ptlrpc_req_finished(request);
554 GOTO(out_lock_cn_cb, err);
557 LASSERT(fid_is_sane(&sbi->ll_root_fid));
558 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
559 sbi->ll_flags & LL_SBI_32BIT_API),
561 md_free_lustre_md(sbi->ll_md_exp, &lmd);
562 ptlrpc_req_finished(request);
565 #ifdef CONFIG_FS_POSIX_ACL
567 posix_acl_release(lmd.posix_acl);
568 lmd.posix_acl = NULL;
571 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
573 CERROR("lustre_lite: bad iget4 for root\n");
577 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
578 if (sbi->ll_checksum_set) {
579 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
580 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
581 sizeof(checksum), &checksum, NULL);
583 CERROR("%s: Set checksum failed: rc = %d\n",
584 sbi->ll_dt_exp->exp_obd->obd_name, err);
590 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
591 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
592 sbi->ll_cache, NULL);
594 CERROR("%s: Set cache_set failed: rc = %d\n",
595 sbi->ll_dt_exp->exp_obd->obd_name, err);
599 sb->s_root = d_make_root(root);
600 if (sb->s_root == NULL) {
601 CERROR("%s: can't make root dentry\n",
602 ll_get_fsname(sb, NULL, 0));
603 GOTO(out_root, err = -ENOMEM);
605 #ifdef HAVE_DCACHE_LOCK
606 sb->s_root->d_op = &ll_d_ops;
609 sbi->ll_sdev_orig = sb->s_dev;
611 /* We set sb->s_dev equal on all lustre clients in order to support
612 * NFS export clustering. NFSD requires that the FSID be the same
614 /* s_dev is also used in lt_compare() to compare two fs, but that is
615 * only a node-local comparison. */
616 uuid = obd_get_uuid(sbi->ll_md_exp);
618 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
625 if (sbi->ll_proc_root != NULL) {
626 err = lprocfs_ll_register_obd(sb, dt);
628 CERROR("%s: could not register %s in llite: rc = %d\n",
629 dt, ll_get_fsname(sb, NULL, 0), err);
632 err = lprocfs_ll_register_obd(sb, md);
634 CERROR("%s: could not register %s in llite: rc = %d\n",
635 md, ll_get_fsname(sb, NULL, 0), err);
645 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
647 obd_disconnect(sbi->ll_dt_exp);
648 sbi->ll_dt_exp = NULL;
650 obd_fid_fini(sbi->ll_md_exp->exp_obd);
652 obd_disconnect(sbi->ll_md_exp);
653 sbi->ll_md_exp = NULL;
662 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
666 size = sizeof(*lmmsize);
667 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
668 KEY_MAX_EASIZE, &size, lmmsize);
670 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
671 sbi->ll_dt_exp->exp_obd->obd_name, rc);
676 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
677 KEY_MAX_EASIZE, &size, lmmsize);
679 CERROR("Get max mdsize error rc %d\n", rc);
685 * Get the value of the default_easize parameter.
687 * \see client_obd::cl_default_mds_easize
689 * \param[in] sbi superblock info for this filesystem
690 * \param[out] lmmsize pointer to storage location for value
692 * \retval 0 on success
693 * \retval negative negated errno on failure
695 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
700 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
701 KEY_DEFAULT_EASIZE, &size, lmmsize);
703 CERROR("Get default mdsize error rc %d\n", rc);
709 * Set the default_easize parameter to the given value.
711 * \see client_obd::cl_default_mds_easize
713 * \param[in] sbi superblock info for this filesystem
714 * \param[in] lmmsize the size to set
716 * \retval 0 on success
717 * \retval negative negated errno on failure
719 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
723 if (lmmsize < sizeof(struct lov_mds_md) ||
724 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
727 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
728 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
729 sizeof(int), &lmmsize, NULL);
734 static void client_common_put_super(struct super_block *sb)
736 struct ll_sb_info *sbi = ll_s2sbi(sb);
741 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
742 obd_disconnect(sbi->ll_dt_exp);
743 sbi->ll_dt_exp = NULL;
745 ll_debugfs_unregister_super(sb);
747 obd_fid_fini(sbi->ll_md_exp->exp_obd);
748 obd_disconnect(sbi->ll_md_exp);
749 sbi->ll_md_exp = NULL;
754 void ll_kill_super(struct super_block *sb)
756 struct ll_sb_info *sbi;
760 if (!(sb->s_flags & MS_ACTIVE))
764 /* we need restore s_dev from changed for clustred NFS before put_super
765 * because new kernels have cached s_dev and change sb->s_dev in
766 * put_super not affected real removing devices */
768 sb->s_dev = sbi->ll_sdev_orig;
769 sbi->ll_umounting = 1;
771 /* wait running statahead threads to quit */
772 while (atomic_read(&sbi->ll_sa_running) > 0) {
773 set_current_state(TASK_UNINTERRUPTIBLE);
774 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
781 static inline int ll_set_opt(const char *opt, char *data, int fl)
783 if (strncmp(opt, data, strlen(opt)) != 0)
789 /* non-client-specific mount options are parsed in lmd_parse */
790 static int ll_options(char *options, struct ll_sb_info *sbi)
793 char *s1 = options, *s2;
794 int *flags = &sbi->ll_flags;
800 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
803 CDEBUG(D_SUPER, "next opt=%s\n", s1);
804 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
809 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
814 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
819 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
824 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
829 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
834 tmp = ll_set_opt("context", s1, 1);
837 tmp = ll_set_opt("fscontext", s1, 1);
840 tmp = ll_set_opt("defcontext", s1, 1);
843 tmp = ll_set_opt("rootcontext", s1, 1);
846 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
851 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
857 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
860 sbi->ll_checksum_set = 1;
863 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
866 sbi->ll_checksum_set = 1;
869 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
874 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
879 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
884 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
889 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
894 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
899 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
904 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
909 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
915 s2 = strchr(s1, ',');
923 void ll_lli_init(struct ll_inode_info *lli)
925 lli->lli_inode_magic = LLI_INODE_MAGIC;
927 spin_lock_init(&lli->lli_lock);
928 lli->lli_posix_acl = NULL;
929 /* Do not set lli_fid, it has been initialized already. */
930 fid_zero(&lli->lli_pfid);
931 lli->lli_mds_read_och = NULL;
932 lli->lli_mds_write_och = NULL;
933 lli->lli_mds_exec_och = NULL;
934 lli->lli_open_fd_read_count = 0;
935 lli->lli_open_fd_write_count = 0;
936 lli->lli_open_fd_exec_count = 0;
937 mutex_init(&lli->lli_och_mutex);
938 spin_lock_init(&lli->lli_agl_lock);
939 spin_lock_init(&lli->lli_layout_lock);
940 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
941 lli->lli_clob = NULL;
943 init_rwsem(&lli->lli_xattrs_list_rwsem);
944 mutex_init(&lli->lli_xattrs_enq_lock);
946 LASSERT(lli->lli_vfs_inode.i_mode != 0);
947 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
948 mutex_init(&lli->lli_readdir_mutex);
949 lli->lli_opendir_key = NULL;
951 spin_lock_init(&lli->lli_sa_lock);
952 lli->lli_opendir_pid = 0;
953 lli->lli_sa_enabled = 0;
954 lli->lli_def_stripe_offset = -1;
956 mutex_init(&lli->lli_size_mutex);
957 lli->lli_symlink_name = NULL;
958 init_rwsem(&lli->lli_trunc_sem);
959 range_lock_tree_init(&lli->lli_write_tree);
960 init_rwsem(&lli->lli_glimpse_sem);
961 lli->lli_glimpse_time = ktime_set(0, 0);
962 INIT_LIST_HEAD(&lli->lli_agl_list);
963 lli->lli_agl_index = 0;
964 lli->lli_async_rc = 0;
966 mutex_init(&lli->lli_layout_mutex);
967 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
970 #define MAX_STRING_SIZE 128
972 #ifndef HAVE_SUPER_SETUP_BDI_NAME
974 #define LSI_BDI_INITIALIZED 0x00400000
976 #ifndef HAVE_BDI_CAP_MAP_COPY
977 # define BDI_CAP_MAP_COPY 0
980 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
982 struct lustre_sb_info *lsi = s2lsi(sb);
983 char buf[MAX_STRING_SIZE];
987 err = bdi_init(&lsi->lsi_bdi);
991 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
992 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
993 lsi->lsi_bdi.name = "lustre";
995 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
997 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1000 sb->s_bdi = &lsi->lsi_bdi;
1004 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1006 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1008 struct lustre_profile *lprof = NULL;
1009 struct lustre_sb_info *lsi = s2lsi(sb);
1010 struct ll_sb_info *sbi = NULL;
1011 char *dt = NULL, *md = NULL;
1012 char *profilenm = get_profile_name(sb);
1013 struct config_llog_instance *cfg;
1014 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1015 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1016 char name[MAX_STRING_SIZE];
1024 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1026 try_module_get(THIS_MODULE);
1030 GOTO(out_free, err = -ENOMEM);
1032 /* client additional sb info */
1033 lsi->lsi_llsbi = sbi = ll_init_sbi();
1035 GOTO(out_free, err = -ENOMEM);
1037 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1039 GOTO(out_free, err);
1041 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1043 GOTO(out_free, err);
1045 #ifndef HAVE_DCACHE_LOCK
1046 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1047 sb->s_d_op = &ll_d_ops;
1050 len = strlen(lsi->lsi_lmd->lmd_profile);
1051 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1052 if (ptr && (strcmp(ptr, "-client") == 0))
1056 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1057 lsi->lsi_lmd->lmd_profile, sb);
1059 /* Call ll_debugfs_register_super() before lustre_process_log()
1060 * so that "llite.*.*" params can be processed correctly.
1062 err = ll_debugfs_register_super(sb, name);
1064 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1065 ll_get_fsname(sb, NULL, 0), err);
1069 /* Generate a string unique to this super, in case some joker tries
1070 to mount the same fs at two mount points.
1071 Use the address of the super itself.*/
1072 cfg->cfg_instance = sb;
1073 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1074 cfg->cfg_callback = class_config_llog_handler;
1075 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1076 /* set up client obds */
1077 err = lustre_process_log(sb, profilenm, cfg);
1079 GOTO(out_proc, err);
1081 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1082 lprof = class_get_profile(profilenm);
1083 if (lprof == NULL) {
1084 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1085 " read from the MGS. Does that filesystem "
1086 "exist?\n", profilenm);
1087 GOTO(out_proc, err = -EINVAL);
1089 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1090 lprof->lp_md, lprof->lp_dt);
1092 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1093 OBD_ALLOC(dt, dt_len);
1095 GOTO(out_proc, err = -ENOMEM);
1096 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1098 md_len = strlen(lprof->lp_md) + instlen + 2;
1099 OBD_ALLOC(md, md_len);
1101 GOTO(out_proc, err = -ENOMEM);
1102 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1104 /* connections, registrations, sb setup */
1105 err = client_common_fill_super(sb, md, dt, mnt);
1107 GOTO(out_proc, err);
1109 sbi->ll_client_common_fill_super_succeeded = 1;
1113 ll_debugfs_unregister_super(sb);
1116 OBD_FREE(md, md_len);
1118 OBD_FREE(dt, dt_len);
1120 class_put_profile(lprof);
1125 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1126 LCONSOLE_WARN("Mounted %s\n", profilenm);
1128 } /* ll_fill_super */
1130 void ll_put_super(struct super_block *sb)
1132 struct config_llog_instance cfg, params_cfg;
1133 struct obd_device *obd;
1134 struct lustre_sb_info *lsi = s2lsi(sb);
1135 struct ll_sb_info *sbi = ll_s2sbi(sb);
1136 char *profilenm = get_profile_name(sb);
1138 int next, force = 1, rc = 0;
1142 GOTO(out_no_sbi, 0);
1144 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1146 cfg.cfg_instance = sb;
1147 lustre_end_log(sb, profilenm, &cfg);
1149 params_cfg.cfg_instance = sb;
1150 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1152 if (sbi->ll_md_exp) {
1153 obd = class_exp2obd(sbi->ll_md_exp);
1155 force = obd->obd_force;
1158 /* Wait for unstable pages to be committed to stable storage */
1160 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1161 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1162 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1166 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1167 if (force == 0 && rc != -EINTR)
1168 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1171 /* We need to set force before the lov_disconnect in
1172 lustre_common_put_super, since l_d cleans up osc's as well. */
1175 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1177 obd->obd_force = force;
1181 if (sbi->ll_client_common_fill_super_succeeded) {
1182 /* Only if client_common_fill_super succeeded */
1183 client_common_put_super(sb);
1187 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1188 class_manual_cleanup(obd);
1191 if (sbi->ll_flags & LL_SBI_VERBOSE)
1192 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1195 class_del_profile(profilenm);
1197 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1198 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1199 bdi_destroy(&lsi->lsi_bdi);
1200 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1205 lsi->lsi_llsbi = NULL;
1207 lustre_common_put_super(sb);
1209 cl_env_cache_purge(~0);
1211 module_put(THIS_MODULE);
1214 } /* client_put_super */
1216 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1218 struct inode *inode = NULL;
1220 /* NOTE: we depend on atomic igrab() -bzzz */
1221 lock_res_and_lock(lock);
1222 if (lock->l_resource->lr_lvb_inode) {
1223 struct ll_inode_info * lli;
1224 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1225 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1226 inode = igrab(lock->l_resource->lr_lvb_inode);
1228 inode = lock->l_resource->lr_lvb_inode;
1229 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1230 D_WARNING, lock, "lr_lvb_inode %p is "
1231 "bogus: magic %08x",
1232 lock->l_resource->lr_lvb_inode,
1233 lli->lli_inode_magic);
1237 unlock_res_and_lock(lock);
1241 void ll_dir_clear_lsm_md(struct inode *inode)
1243 struct ll_inode_info *lli = ll_i2info(inode);
1245 LASSERT(S_ISDIR(inode->i_mode));
1247 if (lli->lli_lsm_md != NULL) {
1248 lmv_free_memmd(lli->lli_lsm_md);
1249 lli->lli_lsm_md = NULL;
1253 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1254 const struct lu_fid *fid,
1255 struct lustre_md *md)
1257 struct ll_sb_info *sbi = ll_s2sbi(sb);
1258 struct mdt_body *body = md->body;
1259 struct inode *inode;
1263 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1264 inode = iget_locked(sb, ino);
1265 if (inode == NULL) {
1266 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1267 ll_get_fsname(sb, NULL, 0), PFID(fid));
1268 RETURN(ERR_PTR(-ENOENT));
1271 if (inode->i_state & I_NEW) {
1272 struct ll_inode_info *lli = ll_i2info(inode);
1273 struct lmv_stripe_md *lsm = md->lmv;
1275 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1276 (body->mbo_mode & S_IFMT);
1277 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1280 LTIME_S(inode->i_mtime) = 0;
1281 LTIME_S(inode->i_atime) = 0;
1282 LTIME_S(inode->i_ctime) = 0;
1285 #ifdef HAVE_BACKING_DEV_INFO
1286 /* initializing backing dev info. */
1287 inode->i_mapping->backing_dev_info =
1288 &s2lsi(inode->i_sb)->lsi_bdi;
1290 inode->i_op = &ll_dir_inode_operations;
1291 inode->i_fop = &ll_dir_operations;
1292 lli->lli_fid = *fid;
1295 LASSERT(lsm != NULL);
1296 /* master object FID */
1297 lli->lli_pfid = body->mbo_fid1;
1298 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1299 lli, PFID(fid), PFID(&lli->lli_pfid));
1300 unlock_new_inode(inode);
1306 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1309 struct lmv_stripe_md *lsm = md->lmv;
1312 LASSERT(lsm != NULL);
1313 /* XXX sigh, this lsm_root initialization should be in
1314 * LMV layer, but it needs ll_iget right now, so we
1315 * put this here right now. */
1316 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1317 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1318 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1319 /* Unfortunately ll_iget will call ll_update_inode,
1320 * where the initialization of slave inode is slightly
1321 * different, so it reset lsm_md to NULL to avoid
1322 * initializing lsm for slave inode. */
1323 lsm->lsm_md_oinfo[i].lmo_root =
1324 ll_iget_anon_dir(inode->i_sb, fid, md);
1325 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1326 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1328 lsm->lsm_md_oinfo[i].lmo_root = NULL;
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)));
1348 * no striped information from request, lustre_md from req does not
1349 * include stripeEA, see ll_md_setattr()
1354 /* Compare the old and new stripe information */
1355 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1356 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1358 bool layout_changed = lsm->lsm_md_layout_version >
1359 old_lsm->lsm_md_layout_version;
1361 int mask = layout_changed ? D_INODE : D_ERROR;
1364 "%s: inode@%p "DFID" lmv layout %s magic %#x/%#x "
1365 "stripe count %d/%d master_mdt %d/%d "
1366 "hash_type %#x/%#x version %d/%d migrate offset %d/%d "
1367 "migrate hash %#x/%#x pool %s/%s\n",
1368 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1369 PFID(&lli->lli_fid),
1370 layout_changed ? "changed" : "mismatch",
1371 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1372 lsm->lsm_md_stripe_count,
1373 old_lsm->lsm_md_stripe_count,
1374 lsm->lsm_md_master_mdt_index,
1375 old_lsm->lsm_md_master_mdt_index,
1376 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1377 lsm->lsm_md_layout_version,
1378 old_lsm->lsm_md_layout_version,
1379 lsm->lsm_md_migrate_offset,
1380 old_lsm->lsm_md_migrate_offset,
1381 lsm->lsm_md_migrate_hash,
1382 old_lsm->lsm_md_migrate_hash,
1383 lsm->lsm_md_pool_name,
1384 old_lsm->lsm_md_pool_name);
1386 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++)
1387 CDEBUG(mask, "old stripe[%d] "DFID"\n",
1388 idx, PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1390 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++)
1391 CDEBUG(mask, "new stripe[%d] "DFID"\n",
1392 idx, PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1394 if (!layout_changed)
1397 ll_dir_clear_lsm_md(inode);
1400 /* set the directory layout */
1401 if (!lli->lli_lsm_md) {
1402 struct cl_attr *attr;
1404 rc = ll_init_lsm_md(inode, md);
1408 /* set md->lmv to NULL, so the following free lustre_md
1409 * will not free this lsm */
1411 lli->lli_lsm_md = lsm;
1413 OBD_ALLOC_PTR(attr);
1417 /* validate the lsm */
1418 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1419 ll_md_blocking_ast);
1425 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1426 md->body->mbo_nlink = attr->cat_nlink;
1427 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1428 md->body->mbo_size = attr->cat_size;
1429 if (md->body->mbo_valid & OBD_MD_FLATIME)
1430 md->body->mbo_atime = attr->cat_atime;
1431 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1432 md->body->mbo_ctime = attr->cat_ctime;
1433 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1434 md->body->mbo_mtime = attr->cat_mtime;
1438 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1439 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1443 /* Compare the old and new stripe information */
1444 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1445 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1448 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1449 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1450 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1451 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1452 inode, lsm, old_lsm,
1453 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1454 lsm->lsm_md_stripe_count,
1455 old_lsm->lsm_md_stripe_count,
1456 lsm->lsm_md_master_mdt_index,
1457 old_lsm->lsm_md_master_mdt_index,
1458 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1459 lsm->lsm_md_layout_version,
1460 old_lsm->lsm_md_layout_version,
1461 lsm->lsm_md_pool_name,
1462 old_lsm->lsm_md_pool_name);
1464 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1465 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1466 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1467 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1470 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1471 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1472 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1473 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1482 void ll_clear_inode(struct inode *inode)
1484 struct ll_inode_info *lli = ll_i2info(inode);
1485 struct ll_sb_info *sbi = ll_i2sbi(inode);
1488 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1489 PFID(ll_inode2fid(inode)), inode);
1491 if (S_ISDIR(inode->i_mode)) {
1492 /* these should have been cleared in ll_file_release */
1493 LASSERT(lli->lli_opendir_key == NULL);
1494 LASSERT(lli->lli_sai == NULL);
1495 LASSERT(lli->lli_opendir_pid == 0);
1498 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1500 LASSERT(!lli->lli_open_fd_write_count);
1501 LASSERT(!lli->lli_open_fd_read_count);
1502 LASSERT(!lli->lli_open_fd_exec_count);
1504 if (lli->lli_mds_write_och)
1505 ll_md_real_close(inode, FMODE_WRITE);
1506 if (lli->lli_mds_exec_och)
1507 ll_md_real_close(inode, FMODE_EXEC);
1508 if (lli->lli_mds_read_och)
1509 ll_md_real_close(inode, FMODE_READ);
1511 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1512 OBD_FREE(lli->lli_symlink_name,
1513 strlen(lli->lli_symlink_name) + 1);
1514 lli->lli_symlink_name = NULL;
1517 ll_xattr_cache_destroy(inode);
1519 #ifdef CONFIG_FS_POSIX_ACL
1520 forget_all_cached_acls(inode);
1521 if (lli->lli_posix_acl) {
1522 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1523 posix_acl_release(lli->lli_posix_acl);
1524 lli->lli_posix_acl = NULL;
1527 lli->lli_inode_magic = LLI_INODE_DEAD;
1529 if (S_ISDIR(inode->i_mode))
1530 ll_dir_clear_lsm_md(inode);
1531 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1532 LASSERT(list_empty(&lli->lli_agl_list));
1535 * XXX This has to be done before lsm is freed below, because
1536 * cl_object still uses inode lsm.
1538 cl_inode_fini(inode);
1543 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1545 struct lustre_md md;
1546 struct inode *inode = dentry->d_inode;
1547 struct ll_sb_info *sbi = ll_i2sbi(inode);
1548 struct ptlrpc_request *request = NULL;
1552 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1553 LUSTRE_OPC_ANY, NULL);
1554 if (IS_ERR(op_data))
1555 RETURN(PTR_ERR(op_data));
1557 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1559 ptlrpc_req_finished(request);
1560 if (rc == -ENOENT) {
1562 /* Unlinked special device node? Or just a race?
1563 * Pretend we done everything. */
1564 if (!S_ISREG(inode->i_mode) &&
1565 !S_ISDIR(inode->i_mode)) {
1566 ia_valid = op_data->op_attr.ia_valid;
1567 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1568 rc = simple_setattr(dentry, &op_data->op_attr);
1569 op_data->op_attr.ia_valid = ia_valid;
1571 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1572 CERROR("md_setattr fails: rc = %d\n", rc);
1577 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1578 sbi->ll_md_exp, &md);
1580 ptlrpc_req_finished(request);
1584 ia_valid = op_data->op_attr.ia_valid;
1585 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1586 * cache is not cleared yet. */
1587 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1588 if (S_ISREG(inode->i_mode))
1590 rc = simple_setattr(dentry, &op_data->op_attr);
1591 if (S_ISREG(inode->i_mode))
1592 inode_unlock(inode);
1593 op_data->op_attr.ia_valid = ia_valid;
1595 rc = ll_update_inode(inode, &md);
1596 ptlrpc_req_finished(request);
1601 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1602 * object(s) determine the file size and mtime. Otherwise, the MDS will
1603 * keep these values until such a time that objects are allocated for it.
1604 * We do the MDS operations first, as it is checking permissions for us.
1605 * We don't to the MDS RPC if there is nothing that we want to store there,
1606 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1607 * going to do an RPC anyways.
1609 * If we are doing a truncate, we will send the mtime and ctime updates
1610 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1611 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1614 * In case of HSMimport, we only set attr on MDS.
1616 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1618 struct inode *inode = dentry->d_inode;
1619 struct ll_inode_info *lli = ll_i2info(inode);
1620 struct md_op_data *op_data = NULL;
1624 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1625 "valid %x, hsm_import %d\n",
1626 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1627 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1630 if (attr->ia_valid & ATTR_SIZE) {
1631 /* Check new size against VFS/VM file size limit and rlimit */
1632 rc = inode_newsize_ok(inode, attr->ia_size);
1636 /* The maximum Lustre file size is variable, based on the
1637 * OST maximum object size and number of stripes. This
1638 * needs another check in addition to the VFS check above. */
1639 if (attr->ia_size > ll_file_maxbytes(inode)) {
1640 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1641 PFID(&lli->lli_fid), attr->ia_size,
1642 ll_file_maxbytes(inode));
1646 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1649 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1650 if (attr->ia_valid & TIMES_SET_FLAGS) {
1651 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1652 !cfs_capable(CFS_CAP_FOWNER))
1656 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1657 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1658 (attr->ia_valid & ATTR_CTIME)) {
1659 attr->ia_ctime = current_time(inode);
1660 attr->ia_valid |= ATTR_CTIME_SET;
1662 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1663 (attr->ia_valid & ATTR_ATIME)) {
1664 attr->ia_atime = current_time(inode);
1665 attr->ia_valid |= ATTR_ATIME_SET;
1667 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1668 (attr->ia_valid & ATTR_MTIME)) {
1669 attr->ia_mtime = current_time(inode);
1670 attr->ia_valid |= ATTR_MTIME_SET;
1673 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1674 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1675 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1676 (s64)ktime_get_real_seconds());
1678 if (S_ISREG(inode->i_mode)) {
1679 if (attr->ia_valid & ATTR_SIZE)
1680 inode_dio_write_done(inode);
1681 inode_unlock(inode);
1684 /* We always do an MDS RPC, even if we're only changing the size;
1685 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1687 OBD_ALLOC_PTR(op_data);
1688 if (op_data == NULL)
1689 GOTO(out, rc = -ENOMEM);
1691 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1692 /* If we are changing file size, file content is
1693 * modified, flag it. */
1694 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1695 op_data->op_bias |= MDS_DATA_MODIFIED;
1696 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1699 if (attr->ia_valid & ATTR_FILE) {
1700 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1702 if (fd->fd_lease_och)
1703 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1706 op_data->op_attr = *attr;
1708 rc = ll_md_setattr(dentry, op_data);
1712 if (!S_ISREG(inode->i_mode) || hsm_import)
1715 if (attr->ia_valid & (ATTR_SIZE |
1716 ATTR_ATIME | ATTR_ATIME_SET |
1717 ATTR_MTIME | ATTR_MTIME_SET |
1718 ATTR_CTIME | ATTR_CTIME_SET)) {
1719 /* For truncate and utimes sending attributes to OSTs, setting
1720 * mtime/atime to the past will be performed under PW [0:EOF]
1721 * extent lock (new_size:EOF for truncate). It may seem
1722 * excessive to send mtime/atime updates to OSTs when not
1723 * setting times to past, but it is necessary due to possible
1724 * time de-synchronization between MDT inode and OST objects */
1725 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1728 /* If the file was restored, it needs to set dirty flag.
1730 * We've already sent MDS_DATA_MODIFIED flag in
1731 * ll_md_setattr() for truncate. However, the MDT refuses to
1732 * set the HS_DIRTY flag on released files, so we have to set
1733 * it again if the file has been restored. Please check how
1734 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1736 * Please notice that if the file is not released, the previous
1737 * MDS_DATA_MODIFIED has taken effect and usually
1738 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1739 * This way we can save an RPC for common open + trunc
1741 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1742 struct hsm_state_set hss = {
1743 .hss_valid = HSS_SETMASK,
1744 .hss_setmask = HS_DIRTY,
1748 rc2 = ll_hsm_state_set(inode, &hss);
1749 /* truncate and write can happen at the same time, so that
1750 * the file can be set modified even though the file is not
1751 * restored from released state, and ll_hsm_state_set() is
1752 * not applicable for the file, and rc2 < 0 is normal in this
1755 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1756 PFID(ll_inode2fid(inode)), rc2);
1761 if (op_data != NULL)
1762 ll_finish_md_op_data(op_data);
1764 if (S_ISREG(inode->i_mode)) {
1766 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1767 inode_dio_wait(inode);
1768 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1769 * flag. ll_update_inode (called from ll_md_setattr), clears
1770 * inode flags, so there is a gap where S_NOSEC is not set.
1771 * This can cause a writer to take the i_mutex unnecessarily,
1772 * but this is safe to do and should be rare. */
1773 inode_has_no_xattr(inode);
1776 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1777 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1782 int ll_setattr(struct dentry *de, struct iattr *attr)
1784 int mode = de->d_inode->i_mode;
1786 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1787 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1788 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1790 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1791 (ATTR_SIZE|ATTR_MODE)) &&
1792 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1793 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1794 !(attr->ia_mode & S_ISGID))))
1795 attr->ia_valid |= ATTR_FORCE;
1797 if ((attr->ia_valid & ATTR_MODE) &&
1799 !(attr->ia_mode & S_ISUID) &&
1800 !(attr->ia_valid & ATTR_KILL_SUID))
1801 attr->ia_valid |= ATTR_KILL_SUID;
1803 if ((attr->ia_valid & ATTR_MODE) &&
1804 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1805 !(attr->ia_mode & S_ISGID) &&
1806 !(attr->ia_valid & ATTR_KILL_SGID))
1807 attr->ia_valid |= ATTR_KILL_SGID;
1809 /* avoid polluted from ATTR_TIMES_SET,
1810 * projid is not expected to be set here */
1811 attr->ia_valid &= ~MDS_ATTR_PROJID;
1813 return ll_setattr_raw(de, attr, false);
1816 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1819 struct obd_statfs obd_osfs;
1824 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1826 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1828 CERROR("md_statfs fails: rc = %d\n", rc);
1832 osfs->os_type = LL_SUPER_MAGIC;
1834 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1835 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1837 if (osfs->os_state & OS_STATE_SUM)
1840 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1841 flags |= OBD_STATFS_NODELAY;
1843 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1845 CERROR("obd_statfs fails: rc = %d\n", rc);
1849 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1850 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1853 osfs->os_bsize = obd_osfs.os_bsize;
1854 osfs->os_blocks = obd_osfs.os_blocks;
1855 osfs->os_bfree = obd_osfs.os_bfree;
1856 osfs->os_bavail = obd_osfs.os_bavail;
1858 /* If we don't have as many objects free on the OST as inodes
1859 * on the MDS, we reduce the total number of inodes to
1860 * compensate, so that the "inodes in use" number is correct.
1862 if (obd_osfs.os_ffree < osfs->os_ffree) {
1863 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1865 osfs->os_ffree = obd_osfs.os_ffree;
1871 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1873 struct super_block *sb = de->d_sb;
1874 struct obd_statfs osfs;
1875 __u64 fsid = huge_encode_dev(sb->s_dev);
1878 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1879 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1881 /* Some amount of caching on the client is allowed */
1882 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1886 statfs_unpack(sfs, &osfs);
1888 /* We need to downshift for all 32-bit kernels, because we can't
1889 * tell if the kernel is being called via sys_statfs64() or not.
1890 * Stop before overflowing f_bsize - in which case it is better
1891 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1892 if (sizeof(long) < 8) {
1893 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1896 osfs.os_blocks >>= 1;
1897 osfs.os_bfree >>= 1;
1898 osfs.os_bavail >>= 1;
1902 sfs->f_blocks = osfs.os_blocks;
1903 sfs->f_bfree = osfs.os_bfree;
1904 sfs->f_bavail = osfs.os_bavail;
1905 sfs->f_fsid.val[0] = (__u32)fsid;
1906 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1910 void ll_inode_size_lock(struct inode *inode)
1912 struct ll_inode_info *lli;
1914 LASSERT(!S_ISDIR(inode->i_mode));
1916 lli = ll_i2info(inode);
1917 mutex_lock(&lli->lli_size_mutex);
1920 void ll_inode_size_unlock(struct inode *inode)
1922 struct ll_inode_info *lli;
1924 lli = ll_i2info(inode);
1925 mutex_unlock(&lli->lli_size_mutex);
1928 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1930 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1931 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1932 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1934 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1937 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1939 struct ll_inode_info *lli = ll_i2info(inode);
1940 struct mdt_body *body = md->body;
1941 struct ll_sb_info *sbi = ll_i2sbi(inode);
1944 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1945 rc = cl_file_inode_init(inode, md);
1950 if (S_ISDIR(inode->i_mode)) {
1951 rc = ll_update_lsm_md(inode, md);
1956 #ifdef CONFIG_FS_POSIX_ACL
1957 if (body->mbo_valid & OBD_MD_FLACL) {
1958 spin_lock(&lli->lli_lock);
1959 if (lli->lli_posix_acl)
1960 posix_acl_release(lli->lli_posix_acl);
1961 lli->lli_posix_acl = md->posix_acl;
1962 spin_unlock(&lli->lli_lock);
1965 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1966 sbi->ll_flags & LL_SBI_32BIT_API);
1967 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1969 if (body->mbo_valid & OBD_MD_FLATIME) {
1970 if (body->mbo_atime > LTIME_S(inode->i_atime))
1971 LTIME_S(inode->i_atime) = body->mbo_atime;
1972 lli->lli_atime = body->mbo_atime;
1975 if (body->mbo_valid & OBD_MD_FLMTIME) {
1976 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1977 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1978 "to %llu\n", inode->i_ino,
1979 LTIME_S(inode->i_mtime), body->mbo_mtime);
1980 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1982 lli->lli_mtime = body->mbo_mtime;
1985 if (body->mbo_valid & OBD_MD_FLCTIME) {
1986 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1987 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1988 lli->lli_ctime = body->mbo_ctime;
1991 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1992 if (body->mbo_valid & OBD_MD_FLFLAGS)
1993 ll_update_inode_flags(inode, body->mbo_flags);
1994 if (body->mbo_valid & OBD_MD_FLMODE)
1995 inode->i_mode = (inode->i_mode & S_IFMT) |
1996 (body->mbo_mode & ~S_IFMT);
1998 if (body->mbo_valid & OBD_MD_FLTYPE)
1999 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2000 (body->mbo_mode & S_IFMT);
2002 LASSERT(inode->i_mode != 0);
2003 if (S_ISREG(inode->i_mode))
2004 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2005 LL_MAX_BLKSIZE_BITS);
2007 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2009 if (body->mbo_valid & OBD_MD_FLUID)
2010 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2011 if (body->mbo_valid & OBD_MD_FLGID)
2012 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2013 if (body->mbo_valid & OBD_MD_FLPROJID)
2014 lli->lli_projid = body->mbo_projid;
2015 if (body->mbo_valid & OBD_MD_FLNLINK)
2016 set_nlink(inode, body->mbo_nlink);
2017 if (body->mbo_valid & OBD_MD_FLRDEV)
2018 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2020 if (body->mbo_valid & OBD_MD_FLID) {
2021 /* FID shouldn't be changed! */
2022 if (fid_is_sane(&lli->lli_fid)) {
2023 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2024 "Trying to change FID "DFID
2025 " to the "DFID", inode "DFID"(%p)\n",
2026 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2027 PFID(ll_inode2fid(inode)), inode);
2029 lli->lli_fid = body->mbo_fid1;
2033 LASSERT(fid_seq(&lli->lli_fid) != 0);
2035 if (body->mbo_valid & OBD_MD_FLSIZE) {
2036 i_size_write(inode, body->mbo_size);
2038 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2039 PFID(ll_inode2fid(inode)),
2040 (unsigned long long)body->mbo_size);
2042 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2043 inode->i_blocks = body->mbo_blocks;
2046 if (body->mbo_valid & OBD_MD_TSTATE) {
2047 /* Set LLIF_FILE_RESTORING if restore ongoing and
2048 * clear it when done to ensure to start again
2049 * glimpsing updated attrs
2051 if (body->mbo_t_state & MS_RESTORE)
2052 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2054 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2060 int ll_read_inode2(struct inode *inode, void *opaque)
2062 struct lustre_md *md = opaque;
2063 struct ll_inode_info *lli = ll_i2info(inode);
2067 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2068 PFID(&lli->lli_fid), inode);
2070 /* Core attributes from the MDS first. This is a new inode, and
2071 * the VFS doesn't zero times in the core inode so we have to do
2072 * it ourselves. They will be overwritten by either MDS or OST
2073 * attributes - we just need to make sure they aren't newer. */
2074 LTIME_S(inode->i_mtime) = 0;
2075 LTIME_S(inode->i_atime) = 0;
2076 LTIME_S(inode->i_ctime) = 0;
2078 rc = ll_update_inode(inode, md);
2082 /* OIDEBUG(inode); */
2084 #ifdef HAVE_BACKING_DEV_INFO
2085 /* initializing backing dev info. */
2086 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2088 if (S_ISREG(inode->i_mode)) {
2089 struct ll_sb_info *sbi = ll_i2sbi(inode);
2090 inode->i_op = &ll_file_inode_operations;
2091 inode->i_fop = sbi->ll_fop;
2092 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2094 } else if (S_ISDIR(inode->i_mode)) {
2095 inode->i_op = &ll_dir_inode_operations;
2096 inode->i_fop = &ll_dir_operations;
2098 } else if (S_ISLNK(inode->i_mode)) {
2099 inode->i_op = &ll_fast_symlink_inode_operations;
2102 inode->i_op = &ll_special_inode_operations;
2104 init_special_inode(inode, inode->i_mode,
2113 void ll_delete_inode(struct inode *inode)
2115 struct ll_inode_info *lli = ll_i2info(inode);
2118 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2119 /* It is last chance to write out dirty pages,
2120 * otherwise we may lose data while umount */
2121 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2123 truncate_inode_pages_final(&inode->i_data);
2125 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2126 "see https://jira.whamcloud.com/browse/LU-118\n",
2127 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2129 #ifdef HAVE_SBOPS_EVICT_INODE
2130 ll_clear_inode(inode);
2137 int ll_iocontrol(struct inode *inode, struct file *file,
2138 unsigned int cmd, unsigned long arg)
2140 struct ll_sb_info *sbi = ll_i2sbi(inode);
2141 struct ptlrpc_request *req = NULL;
2146 case FS_IOC_GETFLAGS: {
2147 struct mdt_body *body;
2148 struct md_op_data *op_data;
2150 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2151 0, 0, LUSTRE_OPC_ANY,
2153 if (IS_ERR(op_data))
2154 RETURN(PTR_ERR(op_data));
2156 op_data->op_valid = OBD_MD_FLFLAGS;
2157 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2158 ll_finish_md_op_data(op_data);
2160 CERROR("%s: failure inode "DFID": rc = %d\n",
2161 sbi->ll_md_exp->exp_obd->obd_name,
2162 PFID(ll_inode2fid(inode)), rc);
2166 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2168 flags = body->mbo_flags;
2170 ptlrpc_req_finished(req);
2172 RETURN(put_user(flags, (int __user *)arg));
2174 case FS_IOC_SETFLAGS: {
2176 struct md_op_data *op_data;
2177 struct cl_object *obj;
2179 if (get_user(flags, (int __user *)arg))
2182 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2183 LUSTRE_OPC_ANY, NULL);
2184 if (IS_ERR(op_data))
2185 RETURN(PTR_ERR(op_data));
2187 op_data->op_attr_flags = flags;
2188 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2189 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2190 ll_finish_md_op_data(op_data);
2191 ptlrpc_req_finished(req);
2195 ll_update_inode_flags(inode, flags);
2197 obj = ll_i2info(inode)->lli_clob;
2201 OBD_ALLOC_PTR(attr);
2205 attr->ia_valid = ATTR_ATTR_FLAG;
2206 rc = cl_setattr_ost(obj, attr, flags);
2218 int ll_flush_ctx(struct inode *inode)
2220 struct ll_sb_info *sbi = ll_i2sbi(inode);
2222 CDEBUG(D_SEC, "flush context for user %d\n",
2223 from_kuid(&init_user_ns, current_uid()));
2225 obd_set_info_async(NULL, sbi->ll_md_exp,
2226 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2228 obd_set_info_async(NULL, sbi->ll_dt_exp,
2229 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2234 /* umount -f client means force down, don't save state */
2235 void ll_umount_begin(struct super_block *sb)
2237 struct ll_sb_info *sbi = ll_s2sbi(sb);
2238 struct obd_device *obd;
2239 struct obd_ioctl_data *ioc_data;
2240 struct l_wait_info lwi;
2241 wait_queue_head_t waitq;
2244 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2245 sb->s_count, atomic_read(&sb->s_active));
2247 obd = class_exp2obd(sbi->ll_md_exp);
2249 CERROR("Invalid MDC connection handle %#llx\n",
2250 sbi->ll_md_exp->exp_handle.h_cookie);
2256 obd = class_exp2obd(sbi->ll_dt_exp);
2258 CERROR("Invalid LOV connection handle %#llx\n",
2259 sbi->ll_dt_exp->exp_handle.h_cookie);
2265 OBD_ALLOC_PTR(ioc_data);
2267 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2268 sizeof *ioc_data, ioc_data, NULL);
2270 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2271 sizeof *ioc_data, ioc_data, NULL);
2273 OBD_FREE_PTR(ioc_data);
2276 /* Really, we'd like to wait until there are no requests outstanding,
2277 * and then continue. For now, we just periodically checking for vfs
2278 * to decrement mnt_cnt and hope to finish it within 10sec.
2280 init_waitqueue_head(&waitq);
2281 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2282 cfs_time_seconds(1), NULL, NULL);
2283 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2288 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2290 struct ll_sb_info *sbi = ll_s2sbi(sb);
2291 char *profilenm = get_profile_name(sb);
2295 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2296 read_only = *flags & MS_RDONLY;
2297 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2298 sizeof(KEY_READ_ONLY),
2299 KEY_READ_ONLY, sizeof(read_only),
2302 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2303 profilenm, read_only ?
2304 "read-only" : "read-write", err);
2309 sb->s_flags |= MS_RDONLY;
2311 sb->s_flags &= ~MS_RDONLY;
2313 if (sbi->ll_flags & LL_SBI_VERBOSE)
2314 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2315 read_only ? "read-only" : "read-write");
2321 * Cleanup the open handle that is cached on MDT-side.
2323 * For open case, the client side open handling thread may hit error
2324 * after the MDT grant the open. Under such case, the client should
2325 * send close RPC to the MDT as cleanup; otherwise, the open handle
2326 * on the MDT will be leaked there until the client umount or evicted.
2328 * In further, if someone unlinked the file, because the open handle
2329 * holds the reference on such file/object, then it will block the
2330 * subsequent threads that want to locate such object via FID.
2332 * \param[in] sb super block for this file-system
2333 * \param[in] open_req pointer to the original open request
2335 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2337 struct mdt_body *body;
2338 struct md_op_data *op_data;
2339 struct ptlrpc_request *close_req = NULL;
2340 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2343 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2344 OBD_ALLOC_PTR(op_data);
2345 if (op_data == NULL) {
2346 CWARN("%s: cannot allocate op_data to release open handle for "
2348 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2353 op_data->op_fid1 = body->mbo_fid1;
2354 op_data->op_open_handle = body->mbo_open_handle;
2355 op_data->op_mod_time = ktime_get_real_seconds();
2356 md_close(exp, op_data, NULL, &close_req);
2357 ptlrpc_req_finished(close_req);
2358 ll_finish_md_op_data(op_data);
2363 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2364 struct super_block *sb, struct lookup_intent *it)
2366 struct ll_sb_info *sbi = NULL;
2367 struct lustre_md md = { NULL };
2371 LASSERT(*inode || sb);
2372 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2373 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2374 sbi->ll_md_exp, &md);
2379 rc = ll_update_inode(*inode, &md);
2383 LASSERT(sb != NULL);
2386 * At this point server returns to client's same fid as client
2387 * generated for creating. So using ->fid1 is okay here.
2389 if (!fid_is_sane(&md.body->mbo_fid1)) {
2390 CERROR("%s: Fid is insane "DFID"\n",
2391 ll_get_fsname(sb, NULL, 0),
2392 PFID(&md.body->mbo_fid1));
2393 GOTO(out, rc = -EINVAL);
2396 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2397 sbi->ll_flags & LL_SBI_32BIT_API),
2399 if (IS_ERR(*inode)) {
2400 #ifdef CONFIG_FS_POSIX_ACL
2402 posix_acl_release(md.posix_acl);
2403 md.posix_acl = NULL;
2406 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2408 CERROR("new_inode -fatal: rc %d\n", rc);
2413 /* Handling piggyback layout lock.
2414 * Layout lock can be piggybacked by getattr and open request.
2415 * The lsm can be applied to inode only if it comes with a layout lock
2416 * otherwise correct layout may be overwritten, for example:
2417 * 1. proc1: mdt returns a lsm but not granting layout
2418 * 2. layout was changed by another client
2419 * 3. proc2: refresh layout and layout lock granted
2420 * 4. proc1: to apply a stale layout */
2421 if (it != NULL && it->it_lock_mode != 0) {
2422 struct lustre_handle lockh;
2423 struct ldlm_lock *lock;
2425 lockh.cookie = it->it_lock_handle;
2426 lock = ldlm_handle2lock(&lockh);
2427 LASSERT(lock != NULL);
2428 if (ldlm_has_layout(lock)) {
2429 struct cl_object_conf conf;
2431 memset(&conf, 0, sizeof(conf));
2432 conf.coc_opc = OBJECT_CONF_SET;
2433 conf.coc_inode = *inode;
2434 conf.coc_lock = lock;
2435 conf.u.coc_layout = md.layout;
2436 (void)ll_layout_conf(*inode, &conf);
2438 LDLM_LOCK_PUT(lock);
2444 md_free_lustre_md(sbi->ll_md_exp, &md);
2447 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2448 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2453 int ll_obd_statfs(struct inode *inode, void __user *arg)
2455 struct ll_sb_info *sbi = NULL;
2456 struct obd_export *exp;
2458 struct obd_ioctl_data *data = NULL;
2462 if (!inode || !(sbi = ll_i2sbi(inode)))
2463 GOTO(out_statfs, rc = -EINVAL);
2465 rc = obd_ioctl_getdata(&buf, &len, arg);
2467 GOTO(out_statfs, rc);
2470 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2471 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2472 GOTO(out_statfs, rc = -EINVAL);
2474 if (data->ioc_inllen1 != sizeof(__u32) ||
2475 data->ioc_inllen2 != sizeof(__u32) ||
2476 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2477 data->ioc_plen2 != sizeof(struct obd_uuid))
2478 GOTO(out_statfs, rc = -EINVAL);
2480 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2481 if (type & LL_STATFS_LMV)
2482 exp = sbi->ll_md_exp;
2483 else if (type & LL_STATFS_LOV)
2484 exp = sbi->ll_dt_exp;
2486 GOTO(out_statfs, rc = -ENODEV);
2488 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2490 GOTO(out_statfs, rc);
2492 OBD_FREE_LARGE(buf, len);
2496 /* this function prepares md_op_data hint for passing it down to MD stack. */
2497 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2498 struct inode *i1, struct inode *i2,
2499 const char *name, size_t namelen,
2500 __u32 mode, __u32 opc, void *data)
2502 LASSERT(i1 != NULL);
2505 /* Do not reuse namelen for something else. */
2507 return ERR_PTR(-EINVAL);
2509 if (namelen > ll_i2sbi(i1)->ll_namelen)
2510 return ERR_PTR(-ENAMETOOLONG);
2512 if (!lu_name_is_valid_2(name, namelen))
2513 return ERR_PTR(-EINVAL);
2516 if (op_data == NULL)
2517 OBD_ALLOC_PTR(op_data);
2519 if (op_data == NULL)
2520 return ERR_PTR(-ENOMEM);
2522 ll_i2gids(op_data->op_suppgids, i1, i2);
2523 op_data->op_fid1 = *ll_inode2fid(i1);
2524 op_data->op_default_stripe_offset = -1;
2525 if (S_ISDIR(i1->i_mode)) {
2526 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2527 if (opc == LUSTRE_OPC_MKDIR)
2528 op_data->op_default_stripe_offset =
2529 ll_i2info(i1)->lli_def_stripe_offset;
2533 op_data->op_fid2 = *ll_inode2fid(i2);
2534 if (S_ISDIR(i2->i_mode))
2535 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2537 fid_zero(&op_data->op_fid2);
2540 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2541 op_data->op_cli_flags |= CLI_HASH64;
2543 if (ll_need_32bit_api(ll_i2sbi(i1)))
2544 op_data->op_cli_flags |= CLI_API32;
2546 op_data->op_name = name;
2547 op_data->op_namelen = namelen;
2548 op_data->op_mode = mode;
2549 op_data->op_mod_time = ktime_get_real_seconds();
2550 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2551 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2552 op_data->op_cap = cfs_curproc_cap_pack();
2553 op_data->op_mds = 0;
2554 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2555 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2556 op_data->op_bias |= MDS_CREATE_VOLATILE;
2558 op_data->op_data = data;
2563 void ll_finish_md_op_data(struct md_op_data *op_data)
2565 security_release_secctx(op_data->op_file_secctx,
2566 op_data->op_file_secctx_size);
2567 OBD_FREE_PTR(op_data);
2570 #ifdef HAVE_SUPEROPS_USE_DENTRY
2571 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2573 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2576 struct ll_sb_info *sbi;
2578 #ifdef HAVE_SUPEROPS_USE_DENTRY
2579 LASSERT((seq != NULL) && (dentry != NULL));
2580 sbi = ll_s2sbi(dentry->d_sb);
2582 LASSERT((seq != NULL) && (vfs != NULL));
2583 sbi = ll_s2sbi(vfs->mnt_sb);
2586 if (sbi->ll_flags & LL_SBI_NOLCK)
2587 seq_puts(seq, ",nolock");
2589 if (sbi->ll_flags & LL_SBI_FLOCK)
2590 seq_puts(seq, ",flock");
2592 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2593 seq_puts(seq, ",localflock");
2595 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2596 seq_puts(seq, ",user_xattr");
2598 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2599 seq_puts(seq, ",lazystatfs");
2601 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2602 seq_puts(seq, ",user_fid2path");
2604 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2605 seq_puts(seq, ",always_ping");
2611 * Get obd name by cmd, and copy out to user space
2613 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2615 struct ll_sb_info *sbi = ll_i2sbi(inode);
2616 struct obd_device *obd;
2619 if (cmd == OBD_IOC_GETDTNAME)
2620 obd = class_exp2obd(sbi->ll_dt_exp);
2621 else if (cmd == OBD_IOC_GETMDNAME)
2622 obd = class_exp2obd(sbi->ll_md_exp);
2629 if (copy_to_user((void __user *)arg, obd->obd_name,
2630 strlen(obd->obd_name) + 1))
2637 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2638 * fsname will be returned in this buffer; otherwise, a static buffer will be
2639 * used to store the fsname and returned to caller.
2641 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2643 static char fsname_static[MTI_NAME_MAXLEN];
2644 struct lustre_sb_info *lsi = s2lsi(sb);
2649 /* this means the caller wants to use static buffer
2650 * and it doesn't care about race. Usually this is
2651 * in error reporting path */
2652 buf = fsname_static;
2653 buflen = sizeof(fsname_static);
2656 len = strlen(lsi->lsi_lmd->lmd_profile);
2657 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2658 if (ptr && (strcmp(ptr, "-client") == 0))
2661 if (unlikely(len >= buflen))
2663 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2669 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2676 p.mnt = current->fs->root.mnt;
2678 path = d_path(&p, buf, bufsize);
2683 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2685 char *buf, *path = NULL;
2686 struct dentry *dentry = NULL;
2687 struct inode *inode = page->mapping->host;
2689 /* this can be called inside spin lock so use GFP_ATOMIC. */
2690 buf = (char *)__get_free_page(GFP_ATOMIC);
2692 dentry = d_find_alias(page->mapping->host);
2694 path = ll_d_path(dentry, buf, PAGE_SIZE);
2698 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2699 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2700 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2701 PFID(ll_inode2fid(inode)),
2702 (path && !IS_ERR(path)) ? path : "", ioret);
2708 free_page((unsigned long)buf);
2711 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2712 struct lov_user_md **kbuf)
2714 struct lov_user_md lum;
2718 if (copy_from_user(&lum, md, sizeof(lum)))
2721 lum_size = ll_lov_user_md_size(&lum);
2725 OBD_ALLOC(*kbuf, lum_size);
2729 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2730 OBD_FREE(*kbuf, lum_size);
2738 * Compute llite root squash state after a change of root squash
2739 * configuration setting or add/remove of a lnet nid
2741 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2743 struct root_squash_info *squash = &sbi->ll_squash;
2746 struct lnet_process_id id;
2748 /* Update norootsquash flag */
2749 down_write(&squash->rsi_sem);
2750 if (list_empty(&squash->rsi_nosquash_nids))
2751 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2753 /* Do not apply root squash as soon as one of our NIDs is
2754 * in the nosquash_nids list */
2757 while (LNetGetId(i++, &id) != -ENOENT) {
2758 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2760 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2766 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2768 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2770 up_write(&squash->rsi_sem);
2774 * Parse linkea content to extract information about a given hardlink
2776 * \param[in] ldata - Initialized linkea data
2777 * \param[in] linkno - Link identifier
2778 * \param[out] parent_fid - The entry's parent FID
2779 * \param[out] ln - Entry name destination buffer
2781 * \retval 0 on success
2782 * \retval Appropriate negative error code on failure
2784 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2785 struct lu_fid *parent_fid, struct lu_name *ln)
2791 rc = linkea_init_with_rec(ldata);
2795 if (linkno >= ldata->ld_leh->leh_reccount)
2796 /* beyond last link */
2799 linkea_first_entry(ldata);
2800 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2801 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2806 linkea_next_entry(ldata);
2816 * Get parent FID and name of an identified link. Operation is performed for
2817 * a given link number, letting the caller iterate over linkno to list one or
2818 * all links of an entry.
2820 * \param[in] file - File descriptor against which to perform the operation
2821 * \param[in,out] arg - User-filled structure containing the linkno to operate
2822 * on and the available size. It is eventually filled with
2823 * the requested information or left untouched on error
2825 * \retval - 0 on success
2826 * \retval - Appropriate negative error code on failure
2828 int ll_getparent(struct file *file, struct getparent __user *arg)
2830 struct inode *inode = file_inode(file);
2831 struct linkea_data *ldata;
2832 struct lu_buf buf = LU_BUF_NULL;
2834 struct lu_fid parent_fid;
2841 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2842 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2845 if (get_user(name_size, &arg->gp_name_size))
2848 if (get_user(linkno, &arg->gp_linkno))
2851 if (name_size > PATH_MAX)
2854 OBD_ALLOC(ldata, sizeof(*ldata));
2858 rc = linkea_data_new(ldata, &buf);
2860 GOTO(ldata_free, rc);
2862 #ifdef HAVE_XATTR_HANDLER_FLAGS
2863 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2864 buf.lb_len, OBD_MD_FLXATTR);
2866 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2868 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2872 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2876 if (ln.ln_namelen >= name_size)
2877 GOTO(lb_free, rc = -EOVERFLOW);
2879 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2880 GOTO(lb_free, rc = -EFAULT);
2882 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2883 GOTO(lb_free, rc = -EFAULT);
2885 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2886 GOTO(lb_free, rc = -EFAULT);
2891 OBD_FREE(ldata, sizeof(*ldata));