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;
87 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 sbi->ll_cache = cl_cache_init(lru_page_max);
95 if (sbi->ll_cache == NULL) {
96 OBD_FREE(sbi, sizeof(*sbi));
100 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
101 SBI_DEFAULT_READAHEAD_MAX);
102 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
103 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
105 ll_generate_random_uuid(uuid);
106 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
109 sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111 sbi->ll_flags |= LL_SBI_CHECKSUM;
114 sbi->ll_flags |= LL_SBI_FLOCK;
117 #ifdef HAVE_LRU_RESIZE_SUPPORT
118 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
120 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
122 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
123 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
129 /* metadata statahead is enabled by default */
130 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
131 sbi->ll_sa_max = LL_SA_RPC_DEF;
132 atomic_set(&sbi->ll_sa_total, 0);
133 atomic_set(&sbi->ll_sa_wrong, 0);
134 atomic_set(&sbi->ll_sa_running, 0);
135 atomic_set(&sbi->ll_agl_total, 0);
136 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
137 sbi->ll_flags |= LL_SBI_FAST_READ;
138 sbi->ll_flags |= LL_SBI_TINY_WRITE;
141 sbi->ll_squash.rsi_uid = 0;
142 sbi->ll_squash.rsi_gid = 0;
143 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
144 init_rwsem(&sbi->ll_squash.rsi_sem);
149 static void ll_free_sbi(struct super_block *sb)
151 struct ll_sb_info *sbi = ll_s2sbi(sb);
155 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
156 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
157 if (sbi->ll_cache != NULL) {
158 cl_cache_decref(sbi->ll_cache);
159 sbi->ll_cache = NULL;
161 OBD_FREE(sbi, sizeof(*sbi));
166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
167 struct vfsmount *mnt)
169 struct inode *root = NULL;
170 struct ll_sb_info *sbi = ll_s2sbi(sb);
171 struct obd_statfs *osfs = NULL;
172 struct ptlrpc_request *request = NULL;
173 struct obd_connect_data *data = NULL;
174 struct obd_uuid *uuid;
175 struct md_op_data *op_data;
176 struct lustre_md lmd;
178 int size, err, checksum;
181 sbi->ll_md_obd = class_name2obd(md);
182 if (!sbi->ll_md_obd) {
183 CERROR("MD %s: not setup or attached\n", md);
197 /* pass client page size via ocd_grant_blkbits, the server should report
198 * back its backend blocksize for grant calculation purpose */
199 data->ocd_grant_blkbits = PAGE_SHIFT;
201 /* indicate MDT features supported by this client */
202 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
203 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
204 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
205 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
206 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
207 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
208 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
209 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
210 OBD_CONNECT_64BITHASH |
211 OBD_CONNECT_EINPROGRESS |
212 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
213 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
214 OBD_CONNECT_MAX_EASIZE |
215 OBD_CONNECT_FLOCK_DEAD |
216 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
217 OBD_CONNECT_OPEN_BY_FID |
218 OBD_CONNECT_DIR_STRIPE |
219 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
220 OBD_CONNECT_SUBTREE |
221 OBD_CONNECT_MULTIMODRPCS |
222 OBD_CONNECT_GRANT_PARAM |
223 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
225 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
226 OBD_CONNECT2_LOCK_CONVERT |
227 OBD_CONNECT2_DIR_MIGRATE |
228 OBD_CONNECT2_SUM_STATFS |
229 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
231 OBD_CONNECT2_ASYNC_DISCARD;
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 obd_connect_set_secctx(data);
277 #if defined(CONFIG_SECURITY)
278 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
281 data->ocd_brw_size = MD_MAX_BRW_SIZE;
283 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
284 &sbi->ll_sb_uuid, data, NULL);
286 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
287 "recovery, of which this client is not a "
288 "part. Please wait for recovery to complete,"
289 " abort, or time out.\n", md);
292 CERROR("cannot connect to %s: rc = %d\n", md, err);
296 sbi->ll_md_exp->exp_connect_data = *data;
298 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
299 LUSTRE_SEQ_METADATA);
301 CERROR("%s: Can't init metadata layer FID infrastructure, "
302 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
306 /* For mount, we only need fs info from MDT0, and also in DNE, it
307 * can make sure the client can be mounted as long as MDT0 is
309 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
310 ktime_get_seconds() - sbi->ll_statfs_max_age,
311 OBD_STATFS_FOR_MDT0);
313 GOTO(out_md_fid, err);
315 /* This needs to be after statfs to ensure connect has finished.
316 * Note that "data" does NOT contain the valid connect reply.
317 * If connecting to a 1.8 server there will be no LMV device, so
318 * we can access the MDC export directly and exp_connect_flags will
319 * be non-zero, but if accessing an upgraded 2.1 server it will
320 * have the correct flags filled in.
321 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
322 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
323 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
324 valid != CLIENT_CONNECT_MDT_REQD) {
327 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
328 obd_connect_flags2str(buf, PAGE_SIZE,
329 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
330 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
331 "feature(s) needed for correct operation "
332 "of this client (%s). Please upgrade "
333 "server or downgrade client.\n",
334 sbi->ll_md_exp->exp_obd->obd_name, buf);
335 OBD_FREE(buf, PAGE_SIZE);
336 GOTO(out_md_fid, err = -EPROTO);
339 size = sizeof(*data);
340 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
341 KEY_CONN_DATA, &size, data);
343 CERROR("%s: Get connect data failed: rc = %d\n",
344 sbi->ll_md_exp->exp_obd->obd_name, err);
345 GOTO(out_md_fid, err);
348 LASSERT(osfs->os_bsize);
349 sb->s_blocksize = osfs->os_bsize;
350 sb->s_blocksize_bits = log2(osfs->os_bsize);
351 sb->s_magic = LL_SUPER_MAGIC;
352 sb->s_maxbytes = MAX_LFS_FILESIZE;
353 sbi->ll_namelen = osfs->os_namelen;
354 sbi->ll_mnt.mnt = current->fs->root.mnt;
356 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
357 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
358 LCONSOLE_INFO("Disabling user_xattr feature because "
359 "it is not supported on the server\n");
360 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
363 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
365 sb->s_flags |= MS_POSIXACL;
367 sbi->ll_flags |= LL_SBI_ACL;
369 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
371 sb->s_flags &= ~MS_POSIXACL;
373 sbi->ll_flags &= ~LL_SBI_ACL;
376 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
377 sbi->ll_flags |= LL_SBI_64BIT_HASH;
379 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
380 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
382 if (obd_connect_has_secctx(data))
383 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
385 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
386 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
387 LCONSOLE_INFO("%s: disabling xattr cache due to "
388 "unknown maximum xattr size.\n", dt);
389 } else if (!sbi->ll_xattr_cache_set) {
390 /* If xattr_cache is already set (no matter 0 or 1)
391 * during processing llog, it won't be enabled here. */
392 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
393 sbi->ll_xattr_cache_enabled = 1;
397 sbi->ll_dt_obd = class_name2obd(dt);
398 if (!sbi->ll_dt_obd) {
399 CERROR("DT %s: not setup or attached\n", dt);
400 GOTO(out_md_fid, err = -ENODEV);
403 /* pass client page size via ocd_grant_blkbits, the server should report
404 * back its backend blocksize for grant calculation purpose */
405 data->ocd_grant_blkbits = PAGE_SHIFT;
407 /* indicate OST features supported by this client */
408 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
409 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
410 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
411 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
412 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
413 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
414 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
415 OBD_CONNECT_EINPROGRESS |
416 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
417 OBD_CONNECT_LAYOUTLOCK |
418 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
419 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
420 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
422 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
423 * can interoperate with an older version of lockahead which was released prior
424 * to landing in master. This support will be dropped when 2.13 development
425 * starts. At the point, we should not just drop the connect flag (below), we
426 * should also remove the support in the code.
428 * Removing it means a few things:
429 * 1. Remove this section here
430 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
431 * 3. Remove function exp_connect_lockahead_old
432 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
434 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
435 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
438 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
440 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
441 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
443 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
444 * disabled by default, because it can still be enabled on the
445 * fly via /sys. As a consequence, we still need to come to an
446 * agreement on the supported algorithms at connect time
448 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
450 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
451 data->ocd_cksum_types = OBD_CKSUM_ADLER;
453 data->ocd_cksum_types = obd_cksum_types_supported_client();
455 #ifdef HAVE_LRU_RESIZE_SUPPORT
456 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
458 /* always ping even if server suppress_pings */
459 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
460 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
462 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
463 "ocd_grant: %d\n", data->ocd_connect_flags,
464 data->ocd_version, data->ocd_grant);
466 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
467 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
469 data->ocd_brw_size = DT_MAX_BRW_SIZE;
471 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
472 &sbi->ll_sb_uuid, data, NULL);
474 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
475 "recovery, of which this client is not a "
476 "part. Please wait for recovery to "
477 "complete, abort, or time out.\n", dt);
480 CERROR("%s: Cannot connect to %s: rc = %d\n",
481 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
485 sbi->ll_dt_exp->exp_connect_data = *data;
487 /* Don't change value if it was specified in the config log */
488 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
489 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
490 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
491 (data->ocd_brw_size >> PAGE_SHIFT));
492 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
493 sbi->ll_ra_info.ra_max_pages_per_file)
494 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
495 sbi->ll_ra_info.ra_max_pages_per_file;
498 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
499 LUSTRE_SEQ_METADATA);
501 CERROR("%s: Can't init data layer FID infrastructure, "
502 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
506 mutex_lock(&sbi->ll_lco.lco_lock);
507 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
508 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
509 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
510 mutex_unlock(&sbi->ll_lco.lco_lock);
512 fid_zero(&sbi->ll_root_fid);
513 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
516 CERROR("cannot mds_connect: rc = %d\n", err);
517 GOTO(out_lock_cn_cb, err);
519 if (!fid_is_sane(&sbi->ll_root_fid)) {
520 CERROR("%s: Invalid root fid "DFID" during mount\n",
521 sbi->ll_md_exp->exp_obd->obd_name,
522 PFID(&sbi->ll_root_fid));
523 GOTO(out_lock_cn_cb, err = -EINVAL);
525 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
527 sb->s_op = &lustre_super_operations;
528 #ifdef HAVE_XATTR_HANDLER_FLAGS
529 sb->s_xattr = ll_xattr_handlers;
531 #if THREAD_SIZE >= 8192 /*b=17630*/
532 sb->s_export_op = &lustre_export_operations;
536 * XXX: move this to after cbd setup? */
537 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
538 if (sbi->ll_flags & LL_SBI_ACL)
539 valid |= OBD_MD_FLACL;
541 OBD_ALLOC_PTR(op_data);
543 GOTO(out_lock_cn_cb, err = -ENOMEM);
545 op_data->op_fid1 = sbi->ll_root_fid;
546 op_data->op_mode = 0;
547 op_data->op_valid = valid;
549 err = md_getattr(sbi->ll_md_exp, op_data, &request);
551 OBD_FREE_PTR(op_data);
553 CERROR("%s: md_getattr failed for root: rc = %d\n",
554 sbi->ll_md_exp->exp_obd->obd_name, err);
555 GOTO(out_lock_cn_cb, err);
558 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
559 sbi->ll_md_exp, &lmd);
561 CERROR("failed to understand root inode md: rc = %d\n", err);
562 ptlrpc_req_finished(request);
563 GOTO(out_lock_cn_cb, err);
566 LASSERT(fid_is_sane(&sbi->ll_root_fid));
567 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
568 sbi->ll_flags & LL_SBI_32BIT_API),
570 md_free_lustre_md(sbi->ll_md_exp, &lmd);
571 ptlrpc_req_finished(request);
574 #ifdef CONFIG_FS_POSIX_ACL
576 posix_acl_release(lmd.posix_acl);
577 lmd.posix_acl = NULL;
580 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
582 CERROR("lustre_lite: bad iget4 for root\n");
586 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
587 if (sbi->ll_checksum_set) {
588 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
589 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
590 sizeof(checksum), &checksum, NULL);
592 CERROR("%s: Set checksum failed: rc = %d\n",
593 sbi->ll_dt_exp->exp_obd->obd_name, err);
599 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
600 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
601 sbi->ll_cache, NULL);
603 CERROR("%s: Set cache_set failed: rc = %d\n",
604 sbi->ll_dt_exp->exp_obd->obd_name, err);
608 sb->s_root = d_make_root(root);
609 if (sb->s_root == NULL) {
610 CERROR("%s: can't make root dentry\n",
611 ll_get_fsname(sb, NULL, 0));
612 GOTO(out_root, err = -ENOMEM);
614 #ifdef HAVE_DCACHE_LOCK
615 sb->s_root->d_op = &ll_d_ops;
618 sbi->ll_sdev_orig = sb->s_dev;
620 /* We set sb->s_dev equal on all lustre clients in order to support
621 * NFS export clustering. NFSD requires that the FSID be the same
623 /* s_dev is also used in lt_compare() to compare two fs, but that is
624 * only a node-local comparison. */
625 uuid = obd_get_uuid(sbi->ll_md_exp);
627 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
634 if (sbi->ll_dt_obd) {
635 err = sysfs_create_link(&sbi->ll_kset.kobj,
636 &sbi->ll_dt_obd->obd_kset.kobj,
637 sbi->ll_dt_obd->obd_type->typ_name);
639 CERROR("%s: could not register %s in llite: rc = %d\n",
640 dt, ll_get_fsname(sb, NULL, 0), err);
645 if (sbi->ll_md_obd) {
646 err = sysfs_create_link(&sbi->ll_kset.kobj,
647 &sbi->ll_md_obd->obd_kset.kobj,
648 sbi->ll_md_obd->obd_type->typ_name);
650 CERROR("%s: could not register %s in llite: rc = %d\n",
651 md, ll_get_fsname(sb, NULL, 0), err);
661 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
663 obd_disconnect(sbi->ll_dt_exp);
664 sbi->ll_dt_exp = NULL;
665 sbi->ll_dt_obd = NULL;
667 obd_fid_fini(sbi->ll_md_exp->exp_obd);
669 obd_disconnect(sbi->ll_md_exp);
670 sbi->ll_md_exp = NULL;
671 sbi->ll_md_obd = NULL;
680 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
684 size = sizeof(*lmmsize);
685 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
686 KEY_MAX_EASIZE, &size, lmmsize);
688 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
689 sbi->ll_dt_exp->exp_obd->obd_name, rc);
694 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
695 KEY_MAX_EASIZE, &size, lmmsize);
697 CERROR("Get max mdsize error rc %d\n", rc);
703 * Get the value of the default_easize parameter.
705 * \see client_obd::cl_default_mds_easize
707 * \param[in] sbi superblock info for this filesystem
708 * \param[out] lmmsize pointer to storage location for value
710 * \retval 0 on success
711 * \retval negative negated errno on failure
713 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
718 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
719 KEY_DEFAULT_EASIZE, &size, lmmsize);
721 CERROR("Get default mdsize error rc %d\n", rc);
727 * Set the default_easize parameter to the given value.
729 * \see client_obd::cl_default_mds_easize
731 * \param[in] sbi superblock info for this filesystem
732 * \param[in] lmmsize the size to set
734 * \retval 0 on success
735 * \retval negative negated errno on failure
737 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
741 if (lmmsize < sizeof(struct lov_mds_md) ||
742 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
745 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
746 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
747 sizeof(int), &lmmsize, NULL);
752 static void client_common_put_super(struct super_block *sb)
754 struct ll_sb_info *sbi = ll_s2sbi(sb);
759 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
760 obd_disconnect(sbi->ll_dt_exp);
761 sbi->ll_dt_exp = NULL;
763 ll_debugfs_unregister_super(sb);
765 obd_fid_fini(sbi->ll_md_exp->exp_obd);
766 obd_disconnect(sbi->ll_md_exp);
767 sbi->ll_md_exp = NULL;
772 void ll_kill_super(struct super_block *sb)
774 struct ll_sb_info *sbi;
778 if (!(sb->s_flags & MS_ACTIVE))
782 /* we need restore s_dev from changed for clustred NFS before put_super
783 * because new kernels have cached s_dev and change sb->s_dev in
784 * put_super not affected real removing devices */
786 sb->s_dev = sbi->ll_sdev_orig;
787 sbi->ll_umounting = 1;
789 /* wait running statahead threads to quit */
790 while (atomic_read(&sbi->ll_sa_running) > 0) {
791 set_current_state(TASK_UNINTERRUPTIBLE);
792 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
799 static inline int ll_set_opt(const char *opt, char *data, int fl)
801 if (strncmp(opt, data, strlen(opt)) != 0)
807 /* non-client-specific mount options are parsed in lmd_parse */
808 static int ll_options(char *options, struct ll_sb_info *sbi)
811 char *s1 = options, *s2;
812 int *flags = &sbi->ll_flags;
818 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
821 CDEBUG(D_SUPER, "next opt=%s\n", s1);
822 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
827 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
829 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
832 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
834 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
837 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
842 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
847 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
852 tmp = ll_set_opt("context", s1, 1);
855 tmp = ll_set_opt("fscontext", s1, 1);
858 tmp = ll_set_opt("defcontext", s1, 1);
861 tmp = ll_set_opt("rootcontext", s1, 1);
864 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
869 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
875 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
878 sbi->ll_checksum_set = 1;
881 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
884 sbi->ll_checksum_set = 1;
887 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
892 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
897 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
902 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
907 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
912 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
917 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
922 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
927 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
933 s2 = strchr(s1, ',');
941 void ll_lli_init(struct ll_inode_info *lli)
943 lli->lli_inode_magic = LLI_INODE_MAGIC;
945 spin_lock_init(&lli->lli_lock);
946 lli->lli_posix_acl = NULL;
947 /* Do not set lli_fid, it has been initialized already. */
948 fid_zero(&lli->lli_pfid);
949 lli->lli_mds_read_och = NULL;
950 lli->lli_mds_write_och = NULL;
951 lli->lli_mds_exec_och = NULL;
952 lli->lli_open_fd_read_count = 0;
953 lli->lli_open_fd_write_count = 0;
954 lli->lli_open_fd_exec_count = 0;
955 mutex_init(&lli->lli_och_mutex);
956 spin_lock_init(&lli->lli_agl_lock);
957 spin_lock_init(&lli->lli_layout_lock);
958 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
959 lli->lli_clob = NULL;
961 init_rwsem(&lli->lli_xattrs_list_rwsem);
962 mutex_init(&lli->lli_xattrs_enq_lock);
964 LASSERT(lli->lli_vfs_inode.i_mode != 0);
965 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
966 mutex_init(&lli->lli_readdir_mutex);
967 lli->lli_opendir_key = NULL;
969 spin_lock_init(&lli->lli_sa_lock);
970 lli->lli_opendir_pid = 0;
971 lli->lli_sa_enabled = 0;
972 lli->lli_def_stripe_offset = -1;
973 init_rwsem(&lli->lli_lsm_sem);
975 mutex_init(&lli->lli_size_mutex);
976 lli->lli_symlink_name = NULL;
977 init_rwsem(&lli->lli_trunc_sem);
978 range_lock_tree_init(&lli->lli_write_tree);
979 init_rwsem(&lli->lli_glimpse_sem);
980 lli->lli_glimpse_time = ktime_set(0, 0);
981 INIT_LIST_HEAD(&lli->lli_agl_list);
982 lli->lli_agl_index = 0;
983 lli->lli_async_rc = 0;
985 mutex_init(&lli->lli_layout_mutex);
986 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
989 #define MAX_STRING_SIZE 128
991 #ifndef HAVE_SUPER_SETUP_BDI_NAME
993 #define LSI_BDI_INITIALIZED 0x00400000
995 #ifndef HAVE_BDI_CAP_MAP_COPY
996 # define BDI_CAP_MAP_COPY 0
999 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1001 struct lustre_sb_info *lsi = s2lsi(sb);
1002 char buf[MAX_STRING_SIZE];
1006 err = bdi_init(&lsi->lsi_bdi);
1010 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1011 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1012 lsi->lsi_bdi.name = "lustre";
1013 va_start(args, fmt);
1014 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1016 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1019 sb->s_bdi = &lsi->lsi_bdi;
1023 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1025 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1027 struct lustre_profile *lprof = NULL;
1028 struct lustre_sb_info *lsi = s2lsi(sb);
1029 struct ll_sb_info *sbi = NULL;
1030 char *dt = NULL, *md = NULL;
1031 char *profilenm = get_profile_name(sb);
1032 struct config_llog_instance *cfg;
1033 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1034 const int instlen = 16 + 2;
1035 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1036 char name[MAX_STRING_SIZE];
1044 /* for ASLR, to map between cfg_instance and hashed ptr */
1045 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1046 profilenm, cfg_instance, sb);
1048 try_module_get(THIS_MODULE);
1052 GOTO(out_free_cfg, err = -ENOMEM);
1054 /* client additional sb info */
1055 lsi->lsi_llsbi = sbi = ll_init_sbi();
1057 GOTO(out_free_cfg, err = -ENOMEM);
1059 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1061 GOTO(out_free_cfg, err);
1063 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1065 GOTO(out_free_cfg, err);
1067 #ifndef HAVE_DCACHE_LOCK
1068 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1069 sb->s_d_op = &ll_d_ops;
1072 len = strlen(profilenm);
1073 ptr = strrchr(profilenm, '-');
1074 if (ptr && (strcmp(ptr, "-client") == 0))
1078 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1079 profilenm, cfg_instance);
1081 /* Call ll_debugfs_register_super() before lustre_process_log()
1082 * so that "llite.*.*" params can be processed correctly.
1084 err = ll_debugfs_register_super(sb, name);
1086 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1087 ll_get_fsname(sb, NULL, 0), err);
1091 /* The cfg_instance is a value unique to this super, in case some
1092 * joker tries to mount the same fs at two mount points.
1094 cfg->cfg_instance = cfg_instance;
1095 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1096 cfg->cfg_callback = class_config_llog_handler;
1097 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1098 /* set up client obds */
1099 err = lustre_process_log(sb, profilenm, cfg);
1101 GOTO(out_debugfs, err);
1103 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1104 lprof = class_get_profile(profilenm);
1105 if (lprof == NULL) {
1106 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1107 " read from the MGS. Does that filesystem "
1108 "exist?\n", profilenm);
1109 GOTO(out_debugfs, err = -EINVAL);
1111 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1112 lprof->lp_md, lprof->lp_dt);
1114 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1115 OBD_ALLOC(dt, dt_len);
1117 GOTO(out_profile, err = -ENOMEM);
1118 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1120 md_len = strlen(lprof->lp_md) + instlen + 2;
1121 OBD_ALLOC(md, md_len);
1123 GOTO(out_free_dt, err = -ENOMEM);
1124 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1126 /* connections, registrations, sb setup */
1127 err = client_common_fill_super(sb, md, dt, mnt);
1129 GOTO(out_free_md, err);
1131 sbi->ll_client_common_fill_super_succeeded = 1;
1135 OBD_FREE(md, md_len);
1138 OBD_FREE(dt, dt_len);
1141 class_put_profile(lprof);
1144 ll_debugfs_unregister_super(sb);
1151 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1152 LCONSOLE_WARN("Mounted %s\n", profilenm);
1154 } /* ll_fill_super */
1156 void ll_put_super(struct super_block *sb)
1158 struct config_llog_instance cfg, params_cfg;
1159 struct obd_device *obd;
1160 struct lustre_sb_info *lsi = s2lsi(sb);
1161 struct ll_sb_info *sbi = ll_s2sbi(sb);
1162 char *profilenm = get_profile_name(sb);
1163 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1165 int next, force = 1, rc = 0;
1169 GOTO(out_no_sbi, 0);
1171 /* Should replace instance_id with something better for ASLR */
1172 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1173 profilenm, cfg_instance, sb);
1175 cfg.cfg_instance = cfg_instance;
1176 lustre_end_log(sb, profilenm, &cfg);
1178 params_cfg.cfg_instance = cfg_instance;
1179 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1181 if (sbi->ll_md_exp) {
1182 obd = class_exp2obd(sbi->ll_md_exp);
1184 force = obd->obd_force;
1187 /* Wait for unstable pages to be committed to stable storage */
1189 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1190 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1191 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1195 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1196 if (force == 0 && rc != -EINTR)
1197 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1199 /* We need to set force before the lov_disconnect in
1200 lustre_common_put_super, since l_d cleans up osc's as well. */
1203 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1205 obd->obd_force = force;
1209 if (sbi->ll_client_common_fill_super_succeeded) {
1210 /* Only if client_common_fill_super succeeded */
1211 client_common_put_super(sb);
1215 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1216 class_manual_cleanup(obd);
1219 if (sbi->ll_flags & LL_SBI_VERBOSE)
1220 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1223 class_del_profile(profilenm);
1225 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1226 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1227 bdi_destroy(&lsi->lsi_bdi);
1228 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1233 lsi->lsi_llsbi = NULL;
1235 lustre_common_put_super(sb);
1237 cl_env_cache_purge(~0);
1239 module_put(THIS_MODULE);
1242 } /* client_put_super */
1244 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1246 struct inode *inode = NULL;
1248 /* NOTE: we depend on atomic igrab() -bzzz */
1249 lock_res_and_lock(lock);
1250 if (lock->l_resource->lr_lvb_inode) {
1251 struct ll_inode_info * lli;
1252 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1253 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1254 inode = igrab(lock->l_resource->lr_lvb_inode);
1256 inode = lock->l_resource->lr_lvb_inode;
1257 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1258 D_WARNING, lock, "lr_lvb_inode %p is "
1259 "bogus: magic %08x",
1260 lock->l_resource->lr_lvb_inode,
1261 lli->lli_inode_magic);
1265 unlock_res_and_lock(lock);
1269 void ll_dir_clear_lsm_md(struct inode *inode)
1271 struct ll_inode_info *lli = ll_i2info(inode);
1273 LASSERT(S_ISDIR(inode->i_mode));
1275 if (lli->lli_lsm_md != NULL) {
1276 lmv_free_memmd(lli->lli_lsm_md);
1277 lli->lli_lsm_md = NULL;
1281 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1282 const struct lu_fid *fid,
1283 struct lustre_md *md)
1285 struct ll_sb_info *sbi = ll_s2sbi(sb);
1286 struct mdt_body *body = md->body;
1287 struct inode *inode;
1291 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1292 inode = iget_locked(sb, ino);
1293 if (inode == NULL) {
1294 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1295 ll_get_fsname(sb, NULL, 0), PFID(fid));
1296 RETURN(ERR_PTR(-ENOENT));
1299 if (inode->i_state & I_NEW) {
1300 struct ll_inode_info *lli = ll_i2info(inode);
1301 struct lmv_stripe_md *lsm = md->lmv;
1303 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1304 (body->mbo_mode & S_IFMT);
1305 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1308 inode->i_mtime.tv_sec = 0;
1309 inode->i_atime.tv_sec = 0;
1310 inode->i_ctime.tv_sec = 0;
1313 #ifdef HAVE_BACKING_DEV_INFO
1314 /* initializing backing dev info. */
1315 inode->i_mapping->backing_dev_info =
1316 &s2lsi(inode->i_sb)->lsi_bdi;
1318 inode->i_op = &ll_dir_inode_operations;
1319 inode->i_fop = &ll_dir_operations;
1320 lli->lli_fid = *fid;
1323 LASSERT(lsm != NULL);
1324 /* master object FID */
1325 lli->lli_pfid = body->mbo_fid1;
1326 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1327 lli, PFID(fid), PFID(&lli->lli_pfid));
1328 unlock_new_inode(inode);
1334 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1337 struct lmv_stripe_md *lsm = md->lmv;
1338 struct ll_inode_info *lli = ll_i2info(inode);
1341 LASSERT(lsm != NULL);
1343 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1344 ll_get_fsname(inode->i_sb, NULL, 0),
1345 PFID(&lli->lli_fid));
1346 lsm_md_dump(D_INODE, lsm);
1348 /* XXX sigh, this lsm_root initialization should be in
1349 * LMV layer, but it needs ll_iget right now, so we
1350 * put this here right now. */
1351 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1352 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1353 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1355 if (!fid_is_sane(fid))
1358 /* Unfortunately ll_iget will call ll_update_inode,
1359 * where the initialization of slave inode is slightly
1360 * different, so it reset lsm_md to NULL to avoid
1361 * initializing lsm for slave inode. */
1362 lsm->lsm_md_oinfo[i].lmo_root =
1363 ll_iget_anon_dir(inode->i_sb, fid, md);
1364 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1365 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1367 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1369 iput(lsm->lsm_md_oinfo[i].lmo_root);
1370 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1376 lli->lli_lsm_md = lsm;
1381 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1383 struct ll_inode_info *lli = ll_i2info(inode);
1384 struct lmv_stripe_md *lsm = md->lmv;
1385 struct cl_attr *attr;
1390 LASSERT(S_ISDIR(inode->i_mode));
1391 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1392 PFID(ll_inode2fid(inode)));
1395 * no striped information from request, lustre_md from req does not
1396 * include stripeEA, see ll_md_setattr()
1402 * normally dir layout doesn't change, only take read lock to check
1403 * that to avoid blocking other MD operations.
1405 down_read(&lli->lli_lsm_sem);
1407 /* some concurrent lookup initialized lsm, and unchanged */
1408 if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1409 GOTO(unlock, rc = 0);
1411 /* if dir layout doesn't match, check whether version is increased,
1412 * which means layout is changed, this happens in dir split/merge and
1415 if (lli->lli_lsm_md &&
1416 lsm->lsm_md_layout_version <=
1417 lli->lli_lsm_md->lsm_md_layout_version) {
1418 CERROR("%s: "DFID" dir layout mismatch:\n",
1419 ll_get_fsname(inode->i_sb, NULL, 0),
1420 PFID(&lli->lli_fid));
1421 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1422 lsm_md_dump(D_ERROR, lsm);
1423 GOTO(unlock, rc = -EINVAL);
1426 up_read(&lli->lli_lsm_sem);
1427 down_write(&lli->lli_lsm_sem);
1428 /* clear existing lsm */
1429 if (lli->lli_lsm_md) {
1430 lmv_free_memmd(lli->lli_lsm_md);
1431 lli->lli_lsm_md = NULL;
1434 rc = ll_init_lsm_md(inode, md);
1435 up_write(&lli->lli_lsm_sem);
1439 /* set md->lmv to NULL, so the following free lustre_md will not free
1444 /* md_merge_attr() may take long, since lsm is already set, switch to
1447 down_read(&lli->lli_lsm_sem);
1448 OBD_ALLOC_PTR(attr);
1450 GOTO(unlock, rc = -ENOMEM);
1452 /* validate the lsm */
1453 rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1454 ll_md_blocking_ast);
1456 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1457 md->body->mbo_nlink = attr->cat_nlink;
1458 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1459 md->body->mbo_size = attr->cat_size;
1460 if (md->body->mbo_valid & OBD_MD_FLATIME)
1461 md->body->mbo_atime = attr->cat_atime;
1462 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1463 md->body->mbo_ctime = attr->cat_ctime;
1464 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1465 md->body->mbo_mtime = attr->cat_mtime;
1471 up_read(&lli->lli_lsm_sem);
1476 void ll_clear_inode(struct inode *inode)
1478 struct ll_inode_info *lli = ll_i2info(inode);
1479 struct ll_sb_info *sbi = ll_i2sbi(inode);
1482 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1483 PFID(ll_inode2fid(inode)), inode);
1485 if (S_ISDIR(inode->i_mode)) {
1486 /* these should have been cleared in ll_file_release */
1487 LASSERT(lli->lli_opendir_key == NULL);
1488 LASSERT(lli->lli_sai == NULL);
1489 LASSERT(lli->lli_opendir_pid == 0);
1492 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1494 LASSERT(!lli->lli_open_fd_write_count);
1495 LASSERT(!lli->lli_open_fd_read_count);
1496 LASSERT(!lli->lli_open_fd_exec_count);
1498 if (lli->lli_mds_write_och)
1499 ll_md_real_close(inode, FMODE_WRITE);
1500 if (lli->lli_mds_exec_och)
1501 ll_md_real_close(inode, FMODE_EXEC);
1502 if (lli->lli_mds_read_och)
1503 ll_md_real_close(inode, FMODE_READ);
1505 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1506 OBD_FREE(lli->lli_symlink_name,
1507 strlen(lli->lli_symlink_name) + 1);
1508 lli->lli_symlink_name = NULL;
1511 ll_xattr_cache_destroy(inode);
1513 #ifdef CONFIG_FS_POSIX_ACL
1514 forget_all_cached_acls(inode);
1515 if (lli->lli_posix_acl) {
1516 posix_acl_release(lli->lli_posix_acl);
1517 lli->lli_posix_acl = NULL;
1520 lli->lli_inode_magic = LLI_INODE_DEAD;
1522 if (S_ISDIR(inode->i_mode))
1523 ll_dir_clear_lsm_md(inode);
1524 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1525 LASSERT(list_empty(&lli->lli_agl_list));
1528 * XXX This has to be done before lsm is freed below, because
1529 * cl_object still uses inode lsm.
1531 cl_inode_fini(inode);
1536 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1538 struct lustre_md md;
1539 struct inode *inode = dentry->d_inode;
1540 struct ll_sb_info *sbi = ll_i2sbi(inode);
1541 struct ptlrpc_request *request = NULL;
1545 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1546 LUSTRE_OPC_ANY, NULL);
1547 if (IS_ERR(op_data))
1548 RETURN(PTR_ERR(op_data));
1550 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1552 ptlrpc_req_finished(request);
1553 if (rc == -ENOENT) {
1555 /* Unlinked special device node? Or just a race?
1556 * Pretend we done everything. */
1557 if (!S_ISREG(inode->i_mode) &&
1558 !S_ISDIR(inode->i_mode)) {
1559 ia_valid = op_data->op_attr.ia_valid;
1560 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1561 rc = simple_setattr(dentry, &op_data->op_attr);
1562 op_data->op_attr.ia_valid = ia_valid;
1564 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1565 CERROR("md_setattr fails: rc = %d\n", rc);
1570 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1571 sbi->ll_md_exp, &md);
1573 ptlrpc_req_finished(request);
1577 ia_valid = op_data->op_attr.ia_valid;
1578 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1579 * cache is not cleared yet. */
1580 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1581 if (S_ISREG(inode->i_mode))
1583 rc = simple_setattr(dentry, &op_data->op_attr);
1584 if (S_ISREG(inode->i_mode))
1585 inode_unlock(inode);
1586 op_data->op_attr.ia_valid = ia_valid;
1588 rc = ll_update_inode(inode, &md);
1589 ptlrpc_req_finished(request);
1594 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1595 * object(s) determine the file size and mtime. Otherwise, the MDS will
1596 * keep these values until such a time that objects are allocated for it.
1597 * We do the MDS operations first, as it is checking permissions for us.
1598 * We don't to the MDS RPC if there is nothing that we want to store there,
1599 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1600 * going to do an RPC anyways.
1602 * If we are doing a truncate, we will send the mtime and ctime updates
1603 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1604 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1607 * In case of HSMimport, we only set attr on MDS.
1609 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1610 enum op_xvalid xvalid, bool hsm_import)
1612 struct inode *inode = dentry->d_inode;
1613 struct ll_inode_info *lli = ll_i2info(inode);
1614 struct md_op_data *op_data = NULL;
1618 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1619 "valid %x, hsm_import %d\n",
1620 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1621 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1624 if (attr->ia_valid & ATTR_SIZE) {
1625 /* Check new size against VFS/VM file size limit and rlimit */
1626 rc = inode_newsize_ok(inode, attr->ia_size);
1630 /* The maximum Lustre file size is variable, based on the
1631 * OST maximum object size and number of stripes. This
1632 * needs another check in addition to the VFS check above. */
1633 if (attr->ia_size > ll_file_maxbytes(inode)) {
1634 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1635 PFID(&lli->lli_fid), attr->ia_size,
1636 ll_file_maxbytes(inode));
1640 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1643 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1644 if (attr->ia_valid & TIMES_SET_FLAGS) {
1645 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1646 !cfs_capable(CFS_CAP_FOWNER))
1650 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1651 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1652 (attr->ia_valid & ATTR_CTIME)) {
1653 attr->ia_ctime = current_time(inode);
1654 xvalid |= OP_XVALID_CTIME_SET;
1656 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1657 (attr->ia_valid & ATTR_ATIME)) {
1658 attr->ia_atime = current_time(inode);
1659 attr->ia_valid |= ATTR_ATIME_SET;
1661 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1662 (attr->ia_valid & ATTR_MTIME)) {
1663 attr->ia_mtime = current_time(inode);
1664 attr->ia_valid |= ATTR_MTIME_SET;
1667 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1668 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1669 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1670 ktime_get_real_seconds());
1672 if (S_ISREG(inode->i_mode)) {
1673 if (attr->ia_valid & ATTR_SIZE)
1674 inode_dio_write_done(inode);
1675 inode_unlock(inode);
1678 /* We always do an MDS RPC, even if we're only changing the size;
1679 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1681 OBD_ALLOC_PTR(op_data);
1682 if (op_data == NULL)
1683 GOTO(out, rc = -ENOMEM);
1685 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1686 /* If we are changing file size, file content is
1687 * modified, flag it.
1689 xvalid |= OP_XVALID_OWNEROVERRIDE;
1690 op_data->op_bias |= MDS_DATA_MODIFIED;
1691 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1694 if (attr->ia_valid & ATTR_FILE) {
1695 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1697 if (fd->fd_lease_och)
1698 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1701 op_data->op_attr = *attr;
1702 op_data->op_xvalid = xvalid;
1704 rc = ll_md_setattr(dentry, op_data);
1708 if (!S_ISREG(inode->i_mode) || hsm_import)
1711 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1712 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1713 xvalid & OP_XVALID_CTIME_SET) {
1714 /* For truncate and utimes sending attributes to OSTs, setting
1715 * mtime/atime to the past will be performed under PW [0:EOF]
1716 * extent lock (new_size:EOF for truncate). It may seem
1717 * excessive to send mtime/atime updates to OSTs when not
1718 * setting times to past, but it is necessary due to possible
1719 * time de-synchronization between MDT inode and OST objects
1721 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1724 /* If the file was restored, it needs to set dirty flag.
1726 * We've already sent MDS_DATA_MODIFIED flag in
1727 * ll_md_setattr() for truncate. However, the MDT refuses to
1728 * set the HS_DIRTY flag on released files, so we have to set
1729 * it again if the file has been restored. Please check how
1730 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1732 * Please notice that if the file is not released, the previous
1733 * MDS_DATA_MODIFIED has taken effect and usually
1734 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1735 * This way we can save an RPC for common open + trunc
1737 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1738 struct hsm_state_set hss = {
1739 .hss_valid = HSS_SETMASK,
1740 .hss_setmask = HS_DIRTY,
1744 rc2 = ll_hsm_state_set(inode, &hss);
1745 /* truncate and write can happen at the same time, so that
1746 * the file can be set modified even though the file is not
1747 * restored from released state, and ll_hsm_state_set() is
1748 * not applicable for the file, and rc2 < 0 is normal in this
1751 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1752 PFID(ll_inode2fid(inode)), rc2);
1757 if (op_data != NULL)
1758 ll_finish_md_op_data(op_data);
1760 if (S_ISREG(inode->i_mode)) {
1762 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1763 inode_dio_wait(inode);
1764 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1765 * flag. ll_update_inode (called from ll_md_setattr), clears
1766 * inode flags, so there is a gap where S_NOSEC is not set.
1767 * This can cause a writer to take the i_mutex unnecessarily,
1768 * but this is safe to do and should be rare. */
1769 inode_has_no_xattr(inode);
1772 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1773 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1778 int ll_setattr(struct dentry *de, struct iattr *attr)
1780 int mode = de->d_inode->i_mode;
1781 enum op_xvalid xvalid = 0;
1783 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1784 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1785 xvalid |= OP_XVALID_OWNEROVERRIDE;
1787 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1788 (ATTR_SIZE|ATTR_MODE)) &&
1789 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1790 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1791 !(attr->ia_mode & S_ISGID))))
1792 attr->ia_valid |= ATTR_FORCE;
1794 if ((attr->ia_valid & ATTR_MODE) &&
1796 !(attr->ia_mode & S_ISUID) &&
1797 !(attr->ia_valid & ATTR_KILL_SUID))
1798 attr->ia_valid |= ATTR_KILL_SUID;
1800 if ((attr->ia_valid & ATTR_MODE) &&
1801 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1802 !(attr->ia_mode & S_ISGID) &&
1803 !(attr->ia_valid & ATTR_KILL_SGID))
1804 attr->ia_valid |= ATTR_KILL_SGID;
1806 return ll_setattr_raw(de, attr, xvalid, false);
1809 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1812 struct obd_statfs obd_osfs = { 0 };
1817 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1819 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1823 osfs->os_type = LL_SUPER_MAGIC;
1825 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1826 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1828 if (osfs->os_state & OS_STATE_SUM)
1831 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1832 flags |= OBD_STATFS_NODELAY;
1834 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1835 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1838 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1839 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1842 osfs->os_bsize = obd_osfs.os_bsize;
1843 osfs->os_blocks = obd_osfs.os_blocks;
1844 osfs->os_bfree = obd_osfs.os_bfree;
1845 osfs->os_bavail = obd_osfs.os_bavail;
1847 /* If we have _some_ OSTs, but don't have as many free objects on the
1848 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1849 * to compensate, so that the "inodes in use" number is correct.
1850 * This should be kept in sync with lod_statfs() behaviour.
1852 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1853 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1855 osfs->os_ffree = obd_osfs.os_ffree;
1861 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1863 struct super_block *sb = de->d_sb;
1864 struct obd_statfs osfs;
1865 __u64 fsid = huge_encode_dev(sb->s_dev);
1868 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1869 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1871 /* Some amount of caching on the client is allowed */
1872 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1876 statfs_unpack(sfs, &osfs);
1878 /* We need to downshift for all 32-bit kernels, because we can't
1879 * tell if the kernel is being called via sys_statfs64() or not.
1880 * Stop before overflowing f_bsize - in which case it is better
1881 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1882 if (sizeof(long) < 8) {
1883 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1886 osfs.os_blocks >>= 1;
1887 osfs.os_bfree >>= 1;
1888 osfs.os_bavail >>= 1;
1892 sfs->f_blocks = osfs.os_blocks;
1893 sfs->f_bfree = osfs.os_bfree;
1894 sfs->f_bavail = osfs.os_bavail;
1895 sfs->f_fsid.val[0] = (__u32)fsid;
1896 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1900 void ll_inode_size_lock(struct inode *inode)
1902 struct ll_inode_info *lli;
1904 LASSERT(!S_ISDIR(inode->i_mode));
1906 lli = ll_i2info(inode);
1907 mutex_lock(&lli->lli_size_mutex);
1910 void ll_inode_size_unlock(struct inode *inode)
1912 struct ll_inode_info *lli;
1914 lli = ll_i2info(inode);
1915 mutex_unlock(&lli->lli_size_mutex);
1918 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1920 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1921 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1922 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1924 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1927 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1929 struct ll_inode_info *lli = ll_i2info(inode);
1930 struct mdt_body *body = md->body;
1931 struct ll_sb_info *sbi = ll_i2sbi(inode);
1934 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1935 rc = cl_file_inode_init(inode, md);
1940 if (S_ISDIR(inode->i_mode)) {
1941 rc = ll_update_lsm_md(inode, md);
1946 #ifdef CONFIG_FS_POSIX_ACL
1947 if (body->mbo_valid & OBD_MD_FLACL) {
1948 spin_lock(&lli->lli_lock);
1949 if (lli->lli_posix_acl)
1950 posix_acl_release(lli->lli_posix_acl);
1951 lli->lli_posix_acl = md->posix_acl;
1952 spin_unlock(&lli->lli_lock);
1955 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1956 sbi->ll_flags & LL_SBI_32BIT_API);
1957 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1959 if (body->mbo_valid & OBD_MD_FLATIME) {
1960 if (body->mbo_atime > inode->i_atime.tv_sec)
1961 inode->i_atime.tv_sec = body->mbo_atime;
1962 lli->lli_atime = body->mbo_atime;
1965 if (body->mbo_valid & OBD_MD_FLMTIME) {
1966 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1968 "setting ino %lu mtime from %lld to %llu\n",
1969 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1971 inode->i_mtime.tv_sec = body->mbo_mtime;
1973 lli->lli_mtime = body->mbo_mtime;
1976 if (body->mbo_valid & OBD_MD_FLCTIME) {
1977 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1978 inode->i_ctime.tv_sec = body->mbo_ctime;
1979 lli->lli_ctime = body->mbo_ctime;
1982 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1983 if (body->mbo_valid & OBD_MD_FLFLAGS)
1984 ll_update_inode_flags(inode, body->mbo_flags);
1985 if (body->mbo_valid & OBD_MD_FLMODE)
1986 inode->i_mode = (inode->i_mode & S_IFMT) |
1987 (body->mbo_mode & ~S_IFMT);
1989 if (body->mbo_valid & OBD_MD_FLTYPE)
1990 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1991 (body->mbo_mode & S_IFMT);
1993 LASSERT(inode->i_mode != 0);
1994 if (S_ISREG(inode->i_mode))
1995 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1996 LL_MAX_BLKSIZE_BITS);
1998 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2000 if (body->mbo_valid & OBD_MD_FLUID)
2001 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2002 if (body->mbo_valid & OBD_MD_FLGID)
2003 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2004 if (body->mbo_valid & OBD_MD_FLPROJID)
2005 lli->lli_projid = body->mbo_projid;
2006 if (body->mbo_valid & OBD_MD_FLNLINK)
2007 set_nlink(inode, body->mbo_nlink);
2008 if (body->mbo_valid & OBD_MD_FLRDEV)
2009 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2011 if (body->mbo_valid & OBD_MD_FLID) {
2012 /* FID shouldn't be changed! */
2013 if (fid_is_sane(&lli->lli_fid)) {
2014 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2015 "Trying to change FID "DFID
2016 " to the "DFID", inode "DFID"(%p)\n",
2017 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2018 PFID(ll_inode2fid(inode)), inode);
2020 lli->lli_fid = body->mbo_fid1;
2024 LASSERT(fid_seq(&lli->lli_fid) != 0);
2026 if (body->mbo_valid & OBD_MD_FLSIZE) {
2027 i_size_write(inode, body->mbo_size);
2029 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2030 PFID(ll_inode2fid(inode)),
2031 (unsigned long long)body->mbo_size);
2033 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2034 inode->i_blocks = body->mbo_blocks;
2037 if (body->mbo_valid & OBD_MD_TSTATE) {
2038 /* Set LLIF_FILE_RESTORING if restore ongoing and
2039 * clear it when done to ensure to start again
2040 * glimpsing updated attrs
2042 if (body->mbo_t_state & MS_RESTORE)
2043 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2045 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2051 int ll_read_inode2(struct inode *inode, void *opaque)
2053 struct lustre_md *md = opaque;
2054 struct ll_inode_info *lli = ll_i2info(inode);
2058 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2059 PFID(&lli->lli_fid), inode);
2061 /* Core attributes from the MDS first. This is a new inode, and
2062 * the VFS doesn't zero times in the core inode so we have to do
2063 * it ourselves. They will be overwritten by either MDS or OST
2064 * attributes - we just need to make sure they aren't newer.
2066 inode->i_mtime.tv_sec = 0;
2067 inode->i_atime.tv_sec = 0;
2068 inode->i_ctime.tv_sec = 0;
2070 rc = ll_update_inode(inode, md);
2074 /* OIDEBUG(inode); */
2076 #ifdef HAVE_BACKING_DEV_INFO
2077 /* initializing backing dev info. */
2078 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2080 if (S_ISREG(inode->i_mode)) {
2081 struct ll_sb_info *sbi = ll_i2sbi(inode);
2082 inode->i_op = &ll_file_inode_operations;
2083 inode->i_fop = sbi->ll_fop;
2084 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2086 } else if (S_ISDIR(inode->i_mode)) {
2087 inode->i_op = &ll_dir_inode_operations;
2088 inode->i_fop = &ll_dir_operations;
2090 } else if (S_ISLNK(inode->i_mode)) {
2091 inode->i_op = &ll_fast_symlink_inode_operations;
2094 inode->i_op = &ll_special_inode_operations;
2096 init_special_inode(inode, inode->i_mode,
2105 void ll_delete_inode(struct inode *inode)
2107 struct ll_inode_info *lli = ll_i2info(inode);
2108 struct address_space *mapping = &inode->i_data;
2109 unsigned long nrpages;
2112 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2113 /* It is last chance to write out dirty pages,
2114 * otherwise we may lose data while umount.
2116 * If i_nlink is 0 then just discard data. This is safe because
2117 * local inode gets i_nlink 0 from server only for the last
2118 * unlink, so that file is not opened somewhere else
2120 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2121 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2123 truncate_inode_pages_final(mapping);
2125 /* Workaround for LU-118: Note nrpages may not be totally updated when
2126 * truncate_inode_pages() returns, as there can be a page in the process
2127 * of deletion (inside __delete_from_page_cache()) in the specified
2128 * range. Thus mapping->nrpages can be non-zero when this function
2129 * returns even after truncation of the whole mapping. Only do this if
2130 * npages isn't already zero.
2132 nrpages = mapping->nrpages;
2134 xa_lock_irq(&mapping->i_pages);
2135 nrpages = mapping->nrpages;
2136 xa_unlock_irq(&mapping->i_pages);
2137 } /* Workaround end */
2139 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2140 "see https://jira.whamcloud.com/browse/LU-118\n",
2141 ll_get_fsname(inode->i_sb, NULL, 0),
2142 PFID(ll_inode2fid(inode)), inode, nrpages);
2144 #ifdef HAVE_SBOPS_EVICT_INODE
2145 ll_clear_inode(inode);
2152 int ll_iocontrol(struct inode *inode, struct file *file,
2153 unsigned int cmd, unsigned long arg)
2155 struct ll_sb_info *sbi = ll_i2sbi(inode);
2156 struct ptlrpc_request *req = NULL;
2161 case FS_IOC_GETFLAGS: {
2162 struct mdt_body *body;
2163 struct md_op_data *op_data;
2165 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2166 0, 0, LUSTRE_OPC_ANY,
2168 if (IS_ERR(op_data))
2169 RETURN(PTR_ERR(op_data));
2171 op_data->op_valid = OBD_MD_FLFLAGS;
2172 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2173 ll_finish_md_op_data(op_data);
2175 CERROR("%s: failure inode "DFID": rc = %d\n",
2176 sbi->ll_md_exp->exp_obd->obd_name,
2177 PFID(ll_inode2fid(inode)), rc);
2181 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2183 flags = body->mbo_flags;
2185 ptlrpc_req_finished(req);
2187 RETURN(put_user(flags, (int __user *)arg));
2189 case FS_IOC_SETFLAGS: {
2191 struct md_op_data *op_data;
2192 struct cl_object *obj;
2193 struct fsxattr fa = { 0 };
2195 if (get_user(flags, (int __user *)arg))
2198 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2199 if (flags & LUSTRE_PROJINHERIT_FL)
2200 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2202 rc = ll_ioctl_check_project(inode, &fa);
2206 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2207 LUSTRE_OPC_ANY, NULL);
2208 if (IS_ERR(op_data))
2209 RETURN(PTR_ERR(op_data));
2211 op_data->op_attr_flags = flags;
2212 op_data->op_xvalid |= OP_XVALID_FLAGS;
2213 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2214 ll_finish_md_op_data(op_data);
2215 ptlrpc_req_finished(req);
2219 ll_update_inode_flags(inode, flags);
2221 obj = ll_i2info(inode)->lli_clob;
2225 OBD_ALLOC_PTR(attr);
2229 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2241 int ll_flush_ctx(struct inode *inode)
2243 struct ll_sb_info *sbi = ll_i2sbi(inode);
2245 CDEBUG(D_SEC, "flush context for user %d\n",
2246 from_kuid(&init_user_ns, current_uid()));
2248 obd_set_info_async(NULL, sbi->ll_md_exp,
2249 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2251 obd_set_info_async(NULL, sbi->ll_dt_exp,
2252 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2257 /* umount -f client means force down, don't save state */
2258 void ll_umount_begin(struct super_block *sb)
2260 struct ll_sb_info *sbi = ll_s2sbi(sb);
2261 struct obd_device *obd;
2262 struct obd_ioctl_data *ioc_data;
2263 struct l_wait_info lwi;
2264 wait_queue_head_t waitq;
2267 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2268 sb->s_count, atomic_read(&sb->s_active));
2270 obd = class_exp2obd(sbi->ll_md_exp);
2272 CERROR("Invalid MDC connection handle %#llx\n",
2273 sbi->ll_md_exp->exp_handle.h_cookie);
2279 obd = class_exp2obd(sbi->ll_dt_exp);
2281 CERROR("Invalid LOV connection handle %#llx\n",
2282 sbi->ll_dt_exp->exp_handle.h_cookie);
2288 OBD_ALLOC_PTR(ioc_data);
2290 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2291 sizeof *ioc_data, ioc_data, NULL);
2293 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2294 sizeof *ioc_data, ioc_data, NULL);
2296 OBD_FREE_PTR(ioc_data);
2299 /* Really, we'd like to wait until there are no requests outstanding,
2300 * and then continue. For now, we just periodically checking for vfs
2301 * to decrement mnt_cnt and hope to finish it within 10sec.
2303 init_waitqueue_head(&waitq);
2304 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2305 cfs_time_seconds(1), NULL, NULL);
2306 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2311 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2313 struct ll_sb_info *sbi = ll_s2sbi(sb);
2314 char *profilenm = get_profile_name(sb);
2318 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2319 read_only = *flags & MS_RDONLY;
2320 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2321 sizeof(KEY_READ_ONLY),
2322 KEY_READ_ONLY, sizeof(read_only),
2325 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2326 profilenm, read_only ?
2327 "read-only" : "read-write", err);
2332 sb->s_flags |= MS_RDONLY;
2334 sb->s_flags &= ~MS_RDONLY;
2336 if (sbi->ll_flags & LL_SBI_VERBOSE)
2337 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2338 read_only ? "read-only" : "read-write");
2344 * Cleanup the open handle that is cached on MDT-side.
2346 * For open case, the client side open handling thread may hit error
2347 * after the MDT grant the open. Under such case, the client should
2348 * send close RPC to the MDT as cleanup; otherwise, the open handle
2349 * on the MDT will be leaked there until the client umount or evicted.
2351 * In further, if someone unlinked the file, because the open handle
2352 * holds the reference on such file/object, then it will block the
2353 * subsequent threads that want to locate such object via FID.
2355 * \param[in] sb super block for this file-system
2356 * \param[in] open_req pointer to the original open request
2358 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2360 struct mdt_body *body;
2361 struct md_op_data *op_data;
2362 struct ptlrpc_request *close_req = NULL;
2363 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2366 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2367 OBD_ALLOC_PTR(op_data);
2368 if (op_data == NULL) {
2369 CWARN("%s: cannot allocate op_data to release open handle for "
2371 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2376 op_data->op_fid1 = body->mbo_fid1;
2377 op_data->op_open_handle = body->mbo_open_handle;
2378 op_data->op_mod_time = ktime_get_real_seconds();
2379 md_close(exp, op_data, NULL, &close_req);
2380 ptlrpc_req_finished(close_req);
2381 ll_finish_md_op_data(op_data);
2386 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2387 struct super_block *sb, struct lookup_intent *it)
2389 struct ll_sb_info *sbi = NULL;
2390 struct lustre_md md = { NULL };
2394 LASSERT(*inode || sb);
2395 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2396 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2397 sbi->ll_md_exp, &md);
2402 rc = ll_update_inode(*inode, &md);
2406 LASSERT(sb != NULL);
2409 * At this point server returns to client's same fid as client
2410 * generated for creating. So using ->fid1 is okay here.
2412 if (!fid_is_sane(&md.body->mbo_fid1)) {
2413 CERROR("%s: Fid is insane "DFID"\n",
2414 ll_get_fsname(sb, NULL, 0),
2415 PFID(&md.body->mbo_fid1));
2416 GOTO(out, rc = -EINVAL);
2419 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2420 sbi->ll_flags & LL_SBI_32BIT_API),
2422 if (IS_ERR(*inode)) {
2423 #ifdef CONFIG_FS_POSIX_ACL
2425 posix_acl_release(md.posix_acl);
2426 md.posix_acl = NULL;
2429 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2431 CERROR("new_inode -fatal: rc %d\n", rc);
2436 /* Handling piggyback layout lock.
2437 * Layout lock can be piggybacked by getattr and open request.
2438 * The lsm can be applied to inode only if it comes with a layout lock
2439 * otherwise correct layout may be overwritten, for example:
2440 * 1. proc1: mdt returns a lsm but not granting layout
2441 * 2. layout was changed by another client
2442 * 3. proc2: refresh layout and layout lock granted
2443 * 4. proc1: to apply a stale layout */
2444 if (it != NULL && it->it_lock_mode != 0) {
2445 struct lustre_handle lockh;
2446 struct ldlm_lock *lock;
2448 lockh.cookie = it->it_lock_handle;
2449 lock = ldlm_handle2lock(&lockh);
2450 LASSERT(lock != NULL);
2451 if (ldlm_has_layout(lock)) {
2452 struct cl_object_conf conf;
2454 memset(&conf, 0, sizeof(conf));
2455 conf.coc_opc = OBJECT_CONF_SET;
2456 conf.coc_inode = *inode;
2457 conf.coc_lock = lock;
2458 conf.u.coc_layout = md.layout;
2459 (void)ll_layout_conf(*inode, &conf);
2461 LDLM_LOCK_PUT(lock);
2467 md_free_lustre_md(sbi->ll_md_exp, &md);
2470 if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2471 ll_intent_drop_lock(it);
2472 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2478 int ll_obd_statfs(struct inode *inode, void __user *arg)
2480 struct ll_sb_info *sbi = NULL;
2481 struct obd_export *exp;
2483 struct obd_ioctl_data *data = NULL;
2487 if (!inode || !(sbi = ll_i2sbi(inode)))
2488 GOTO(out_statfs, rc = -EINVAL);
2490 rc = obd_ioctl_getdata(&buf, &len, arg);
2492 GOTO(out_statfs, rc);
2495 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2496 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2497 GOTO(out_statfs, rc = -EINVAL);
2499 if (data->ioc_inllen1 != sizeof(__u32) ||
2500 data->ioc_inllen2 != sizeof(__u32) ||
2501 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2502 data->ioc_plen2 != sizeof(struct obd_uuid))
2503 GOTO(out_statfs, rc = -EINVAL);
2505 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2506 if (type & LL_STATFS_LMV)
2507 exp = sbi->ll_md_exp;
2508 else if (type & LL_STATFS_LOV)
2509 exp = sbi->ll_dt_exp;
2511 GOTO(out_statfs, rc = -ENODEV);
2513 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2515 GOTO(out_statfs, rc);
2517 OBD_FREE_LARGE(buf, len);
2522 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2523 * be called early to avoid deadlock.
2525 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2527 if (op_data->op_mea2_sem) {
2528 up_read(op_data->op_mea2_sem);
2529 op_data->op_mea2_sem = NULL;
2532 if (op_data->op_mea1_sem) {
2533 up_read(op_data->op_mea1_sem);
2534 op_data->op_mea1_sem = NULL;
2538 /* this function prepares md_op_data hint for passing it down to MD stack. */
2539 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2540 struct inode *i1, struct inode *i2,
2541 const char *name, size_t namelen,
2542 __u32 mode, __u32 opc, void *data)
2544 LASSERT(i1 != NULL);
2547 /* Do not reuse namelen for something else. */
2549 return ERR_PTR(-EINVAL);
2551 if (namelen > ll_i2sbi(i1)->ll_namelen)
2552 return ERR_PTR(-ENAMETOOLONG);
2554 if (!lu_name_is_valid_2(name, namelen))
2555 return ERR_PTR(-EINVAL);
2558 if (op_data == NULL)
2559 OBD_ALLOC_PTR(op_data);
2561 if (op_data == NULL)
2562 return ERR_PTR(-ENOMEM);
2564 ll_i2gids(op_data->op_suppgids, i1, i2);
2565 op_data->op_fid1 = *ll_inode2fid(i1);
2566 op_data->op_default_stripe_offset = -1;
2568 if (S_ISDIR(i1->i_mode)) {
2569 down_read(&ll_i2info(i1)->lli_lsm_sem);
2570 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2571 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2572 if (opc == LUSTRE_OPC_MKDIR)
2573 op_data->op_default_stripe_offset =
2574 ll_i2info(i1)->lli_def_stripe_offset;
2578 op_data->op_fid2 = *ll_inode2fid(i2);
2579 if (S_ISDIR(i2->i_mode)) {
2581 down_read(&ll_i2info(i2)->lli_lsm_sem);
2582 op_data->op_mea2_sem =
2583 &ll_i2info(i2)->lli_lsm_sem;
2585 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2588 fid_zero(&op_data->op_fid2);
2591 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2592 op_data->op_cli_flags |= CLI_HASH64;
2594 if (ll_need_32bit_api(ll_i2sbi(i1)))
2595 op_data->op_cli_flags |= CLI_API32;
2597 op_data->op_name = name;
2598 op_data->op_namelen = namelen;
2599 op_data->op_mode = mode;
2600 op_data->op_mod_time = ktime_get_real_seconds();
2601 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2602 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2603 op_data->op_cap = cfs_curproc_cap_pack();
2604 op_data->op_mds = 0;
2605 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2606 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2607 op_data->op_bias |= MDS_CREATE_VOLATILE;
2609 op_data->op_data = data;
2614 void ll_finish_md_op_data(struct md_op_data *op_data)
2616 ll_unlock_md_op_lsm(op_data);
2617 security_release_secctx(op_data->op_file_secctx,
2618 op_data->op_file_secctx_size);
2619 OBD_FREE_PTR(op_data);
2622 #ifdef HAVE_SUPEROPS_USE_DENTRY
2623 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2625 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2628 struct ll_sb_info *sbi;
2630 #ifdef HAVE_SUPEROPS_USE_DENTRY
2631 LASSERT((seq != NULL) && (dentry != NULL));
2632 sbi = ll_s2sbi(dentry->d_sb);
2634 LASSERT((seq != NULL) && (vfs != NULL));
2635 sbi = ll_s2sbi(vfs->mnt_sb);
2638 if (sbi->ll_flags & LL_SBI_NOLCK)
2639 seq_puts(seq, ",nolock");
2641 /* "flock" is the default since 2.13, but it wasn't for many years,
2642 * so it is still useful to print this to show it is enabled.
2643 * Start to print "noflock" so it is now clear when flock is disabled.
2645 if (sbi->ll_flags & LL_SBI_FLOCK)
2646 seq_puts(seq, ",flock");
2647 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2648 seq_puts(seq, ",localflock");
2650 seq_puts(seq, ",noflock");
2652 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2653 seq_puts(seq, ",user_xattr");
2655 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2656 seq_puts(seq, ",lazystatfs");
2658 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2659 seq_puts(seq, ",user_fid2path");
2661 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2662 seq_puts(seq, ",always_ping");
2668 * Get obd name by cmd, and copy out to user space
2670 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2672 struct ll_sb_info *sbi = ll_i2sbi(inode);
2673 struct obd_device *obd;
2676 if (cmd == OBD_IOC_GETDTNAME)
2677 obd = class_exp2obd(sbi->ll_dt_exp);
2678 else if (cmd == OBD_IOC_GETMDNAME)
2679 obd = class_exp2obd(sbi->ll_md_exp);
2686 if (copy_to_user((void __user *)arg, obd->obd_name,
2687 strlen(obd->obd_name) + 1))
2694 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2695 * fsname will be returned in this buffer; otherwise, a static buffer will be
2696 * used to store the fsname and returned to caller.
2698 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2700 static char fsname_static[MTI_NAME_MAXLEN];
2701 struct lustre_sb_info *lsi = s2lsi(sb);
2706 /* this means the caller wants to use static buffer
2707 * and it doesn't care about race. Usually this is
2708 * in error reporting path */
2709 buf = fsname_static;
2710 buflen = sizeof(fsname_static);
2713 len = strlen(lsi->lsi_lmd->lmd_profile);
2714 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2715 if (ptr && (strcmp(ptr, "-client") == 0))
2718 if (unlikely(len >= buflen))
2720 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2726 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2733 p.mnt = current->fs->root.mnt;
2735 path = d_path(&p, buf, bufsize);
2740 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2742 char *buf, *path = NULL;
2743 struct dentry *dentry = NULL;
2744 struct inode *inode = page->mapping->host;
2746 /* this can be called inside spin lock so use GFP_ATOMIC. */
2747 buf = (char *)__get_free_page(GFP_ATOMIC);
2749 dentry = d_find_alias(page->mapping->host);
2751 path = ll_d_path(dentry, buf, PAGE_SIZE);
2755 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2756 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2757 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2758 PFID(ll_inode2fid(inode)),
2759 (path && !IS_ERR(path)) ? path : "", ioret);
2765 free_page((unsigned long)buf);
2768 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2769 struct lov_user_md **kbuf)
2771 struct lov_user_md lum;
2775 if (copy_from_user(&lum, md, sizeof(lum)))
2778 lum_size = ll_lov_user_md_size(&lum);
2782 OBD_ALLOC_LARGE(*kbuf, lum_size);
2786 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2787 OBD_FREE_LARGE(*kbuf, lum_size);
2795 * Compute llite root squash state after a change of root squash
2796 * configuration setting or add/remove of a lnet nid
2798 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2800 struct root_squash_info *squash = &sbi->ll_squash;
2803 struct lnet_process_id id;
2805 /* Update norootsquash flag */
2806 down_write(&squash->rsi_sem);
2807 if (list_empty(&squash->rsi_nosquash_nids))
2808 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2810 /* Do not apply root squash as soon as one of our NIDs is
2811 * in the nosquash_nids list */
2814 while (LNetGetId(i++, &id) != -ENOENT) {
2815 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2817 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2823 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2825 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2827 up_write(&squash->rsi_sem);
2831 * Parse linkea content to extract information about a given hardlink
2833 * \param[in] ldata - Initialized linkea data
2834 * \param[in] linkno - Link identifier
2835 * \param[out] parent_fid - The entry's parent FID
2836 * \param[out] ln - Entry name destination buffer
2838 * \retval 0 on success
2839 * \retval Appropriate negative error code on failure
2841 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2842 struct lu_fid *parent_fid, struct lu_name *ln)
2848 rc = linkea_init_with_rec(ldata);
2852 if (linkno >= ldata->ld_leh->leh_reccount)
2853 /* beyond last link */
2856 linkea_first_entry(ldata);
2857 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2858 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2863 linkea_next_entry(ldata);
2873 * Get parent FID and name of an identified link. Operation is performed for
2874 * a given link number, letting the caller iterate over linkno to list one or
2875 * all links of an entry.
2877 * \param[in] file - File descriptor against which to perform the operation
2878 * \param[in,out] arg - User-filled structure containing the linkno to operate
2879 * on and the available size. It is eventually filled with
2880 * the requested information or left untouched on error
2882 * \retval - 0 on success
2883 * \retval - Appropriate negative error code on failure
2885 int ll_getparent(struct file *file, struct getparent __user *arg)
2887 struct inode *inode = file_inode(file);
2888 struct linkea_data *ldata;
2889 struct lu_buf buf = LU_BUF_NULL;
2891 struct lu_fid parent_fid;
2898 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2899 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2902 if (get_user(name_size, &arg->gp_name_size))
2905 if (get_user(linkno, &arg->gp_linkno))
2908 if (name_size > PATH_MAX)
2911 OBD_ALLOC(ldata, sizeof(*ldata));
2915 rc = linkea_data_new(ldata, &buf);
2917 GOTO(ldata_free, rc);
2919 #ifdef HAVE_XATTR_HANDLER_FLAGS
2920 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2921 buf.lb_len, OBD_MD_FLXATTR);
2923 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2925 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2929 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2933 if (ln.ln_namelen >= name_size)
2934 GOTO(lb_free, rc = -EOVERFLOW);
2936 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2937 GOTO(lb_free, rc = -EFAULT);
2939 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2940 GOTO(lb_free, rc = -EFAULT);
2942 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2943 GOTO(lb_free, rc = -EFAULT);
2948 OBD_FREE(ldata, sizeof(*ldata));