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, 2016, 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_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_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 =
103 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
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 #ifdef HAVE_LRU_RESIZE_SUPPORT
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
119 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 /* metadata statahead is enabled by default */
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;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* indicate the features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_ATTRFID |
202 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
203 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
205 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH |
208 OBD_CONNECT_EINPROGRESS |
209 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211 OBD_CONNECT_MAX_EASIZE |
212 OBD_CONNECT_FLOCK_DEAD |
213 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214 OBD_CONNECT_OPEN_BY_FID |
215 OBD_CONNECT_DIR_STRIPE |
216 OBD_CONNECT_BULK_MBITS |
217 OBD_CONNECT_SUBTREE |
218 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
220 data->ocd_connect_flags2 = 0;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
228 OBD_CONNECT_LARGE_ACL;
231 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232 /* flag mdc connection as lightweight, only used for test
233 * purpose, use with care */
234 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
236 data->ocd_ibits_known = MDS_INODELOCK_FULL;
237 data->ocd_version = LUSTRE_VERSION_CODE;
239 if (sb->s_flags & MS_RDONLY)
240 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
242 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
244 if (sbi->ll_flags & LL_SBI_FLOCK)
245 sbi->ll_fop = &ll_file_operations_flock;
246 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
247 sbi->ll_fop = &ll_file_operations;
249 sbi->ll_fop = &ll_file_operations_noflock;
251 /* always ping even if server suppress_pings */
252 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
253 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
255 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
256 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
257 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
259 data->ocd_brw_size = MD_MAX_BRW_SIZE;
261 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
263 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
264 "recovery, of which this client is not a "
265 "part. Please wait for recovery to complete,"
266 " abort, or time out.\n", md);
269 CERROR("cannot connect to %s: rc = %d\n", md, err);
273 sbi->ll_md_exp->exp_connect_data = *data;
275 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
276 LUSTRE_SEQ_METADATA);
278 CERROR("%s: Can't init metadata layer FID infrastructure, "
279 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
283 /* For mount, we only need fs info from MDT0, and also in DNE, it
284 * can make sure the client can be mounted as long as MDT0 is
286 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
287 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
288 OBD_STATFS_FOR_MDT0);
290 GOTO(out_md_fid, err);
292 /* This needs to be after statfs to ensure connect has finished.
293 * Note that "data" does NOT contain the valid connect reply.
294 * If connecting to a 1.8 server there will be no LMV device, so
295 * we can access the MDC export directly and exp_connect_flags will
296 * be non-zero, but if accessing an upgraded 2.1 server it will
297 * have the correct flags filled in.
298 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
299 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
300 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
301 valid != CLIENT_CONNECT_MDT_REQD) {
304 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
305 obd_connect_flags2str(buf, PAGE_SIZE,
306 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
307 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
308 "feature(s) needed for correct operation "
309 "of this client (%s). Please upgrade "
310 "server or downgrade client.\n",
311 sbi->ll_md_exp->exp_obd->obd_name, buf);
312 OBD_FREE(buf, PAGE_SIZE);
313 GOTO(out_md_fid, err = -EPROTO);
316 size = sizeof(*data);
317 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
318 KEY_CONN_DATA, &size, data);
320 CERROR("%s: Get connect data failed: rc = %d\n",
321 sbi->ll_md_exp->exp_obd->obd_name, err);
322 GOTO(out_md_fid, err);
325 LASSERT(osfs->os_bsize);
326 sb->s_blocksize = osfs->os_bsize;
327 sb->s_blocksize_bits = log2(osfs->os_bsize);
328 sb->s_magic = LL_SUPER_MAGIC;
329 sb->s_maxbytes = MAX_LFS_FILESIZE;
330 sbi->ll_namelen = osfs->os_namelen;
331 sbi->ll_mnt.mnt = current->fs->root.mnt;
333 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
334 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
335 LCONSOLE_INFO("Disabling user_xattr feature because "
336 "it is not supported on the server\n");
337 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
340 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
342 sb->s_flags |= MS_POSIXACL;
344 sbi->ll_flags |= LL_SBI_ACL;
346 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
348 sb->s_flags &= ~MS_POSIXACL;
350 sbi->ll_flags &= ~LL_SBI_ACL;
353 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
354 sbi->ll_flags |= LL_SBI_64BIT_HASH;
356 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
357 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
359 if (obd_connect_has_secctx(data))
360 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
362 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
363 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
364 LCONSOLE_INFO("%s: disabling xattr cache due to "
365 "unknown maximum xattr size.\n", dt);
367 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
368 sbi->ll_xattr_cache_enabled = 1;
372 obd = class_name2obd(dt);
374 CERROR("DT %s: not setup or attached\n", dt);
375 GOTO(out_md_fid, err = -ENODEV);
378 /* pass client page size via ocd_grant_blkbits, the server should report
379 * back its backend blocksize for grant calculation purpose */
380 data->ocd_grant_blkbits = PAGE_SHIFT;
382 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
383 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
384 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
385 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
386 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
387 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
388 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
389 OBD_CONNECT_EINPROGRESS |
390 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
391 OBD_CONNECT_LAYOUTLOCK |
392 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
393 OBD_CONNECT_BULK_MBITS;
395 data->ocd_connect_flags2 = 0;
397 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
398 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
400 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
401 * disabled by default, because it can still be enabled on the
402 * fly via /proc. As a consequence, we still need to come to an
403 * agreement on the supported algorithms at connect time */
404 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
406 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
407 data->ocd_cksum_types = OBD_CKSUM_ADLER;
409 data->ocd_cksum_types = cksum_types_supported_client();
411 #ifdef HAVE_LRU_RESIZE_SUPPORT
412 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
414 /* always ping even if server suppress_pings */
415 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
416 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
418 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
419 "ocd_grant: %d\n", data->ocd_connect_flags,
420 data->ocd_version, data->ocd_grant);
422 obd->obd_upcall.onu_owner = &sbi->ll_lco;
423 obd->obd_upcall.onu_upcall = cl_ocd_update;
425 data->ocd_brw_size = DT_MAX_BRW_SIZE;
427 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
430 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
431 "recovery, of which this client is not a "
432 "part. Please wait for recovery to "
433 "complete, abort, or time out.\n", dt);
436 CERROR("%s: Cannot connect to %s: rc = %d\n",
437 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
441 sbi->ll_dt_exp->exp_connect_data = *data;
443 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
444 LUSTRE_SEQ_METADATA);
446 CERROR("%s: Can't init data layer FID infrastructure, "
447 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
451 mutex_lock(&sbi->ll_lco.lco_lock);
452 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
453 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
454 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
455 mutex_unlock(&sbi->ll_lco.lco_lock);
457 fid_zero(&sbi->ll_root_fid);
458 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
461 CERROR("cannot mds_connect: rc = %d\n", err);
462 GOTO(out_lock_cn_cb, err);
464 if (!fid_is_sane(&sbi->ll_root_fid)) {
465 CERROR("%s: Invalid root fid "DFID" during mount\n",
466 sbi->ll_md_exp->exp_obd->obd_name,
467 PFID(&sbi->ll_root_fid));
468 GOTO(out_lock_cn_cb, err = -EINVAL);
470 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
472 sb->s_op = &lustre_super_operations;
473 #if THREAD_SIZE >= 8192 /*b=17630*/
474 sb->s_export_op = &lustre_export_operations;
478 * XXX: move this to after cbd setup? */
479 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
480 if (sbi->ll_flags & LL_SBI_ACL)
481 valid |= OBD_MD_FLACL;
483 OBD_ALLOC_PTR(op_data);
485 GOTO(out_lock_cn_cb, err = -ENOMEM);
487 op_data->op_fid1 = sbi->ll_root_fid;
488 op_data->op_mode = 0;
489 op_data->op_valid = valid;
491 err = md_getattr(sbi->ll_md_exp, op_data, &request);
493 OBD_FREE_PTR(op_data);
495 CERROR("%s: md_getattr failed for root: rc = %d\n",
496 sbi->ll_md_exp->exp_obd->obd_name, err);
497 GOTO(out_lock_cn_cb, err);
500 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
501 sbi->ll_md_exp, &lmd);
503 CERROR("failed to understand root inode md: rc = %d\n", err);
504 ptlrpc_req_finished(request);
505 GOTO(out_lock_cn_cb, err);
508 LASSERT(fid_is_sane(&sbi->ll_root_fid));
509 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
510 sbi->ll_flags & LL_SBI_32BIT_API),
512 md_free_lustre_md(sbi->ll_md_exp, &lmd);
513 ptlrpc_req_finished(request);
516 #ifdef CONFIG_FS_POSIX_ACL
518 posix_acl_release(lmd.posix_acl);
519 lmd.posix_acl = NULL;
522 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
524 CERROR("lustre_lite: bad iget4 for root\n");
528 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
529 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
530 KEY_CHECKSUM, sizeof(checksum), &checksum,
533 CERROR("%s: Set checksum failed: rc = %d\n",
534 sbi->ll_dt_exp->exp_obd->obd_name, err);
539 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
540 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
541 sbi->ll_cache, NULL);
543 CERROR("%s: Set cache_set failed: rc = %d\n",
544 sbi->ll_dt_exp->exp_obd->obd_name, err);
548 sb->s_root = d_make_root(root);
549 if (sb->s_root == NULL) {
550 CERROR("%s: can't make root dentry\n",
551 ll_get_fsname(sb, NULL, 0));
552 GOTO(out_root, err = -ENOMEM);
554 #ifdef HAVE_DCACHE_LOCK
555 sb->s_root->d_op = &ll_d_ops;
558 sbi->ll_sdev_orig = sb->s_dev;
560 /* We set sb->s_dev equal on all lustre clients in order to support
561 * NFS export clustering. NFSD requires that the FSID be the same
563 /* s_dev is also used in lt_compare() to compare two fs, but that is
564 * only a node-local comparison. */
565 uuid = obd_get_uuid(sbi->ll_md_exp);
567 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
573 if (proc_lustre_fs_root != NULL) {
574 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
577 CERROR("%s: could not register mount in lprocfs: "
578 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
588 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
590 obd_disconnect(sbi->ll_dt_exp);
591 sbi->ll_dt_exp = NULL;
593 obd_fid_fini(sbi->ll_md_exp->exp_obd);
595 obd_disconnect(sbi->ll_md_exp);
596 sbi->ll_md_exp = NULL;
605 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
609 size = sizeof(*lmmsize);
610 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
611 KEY_MAX_EASIZE, &size, lmmsize);
613 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
614 sbi->ll_dt_exp->exp_obd->obd_name, rc);
619 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
620 KEY_MAX_EASIZE, &size, lmmsize);
622 CERROR("Get max mdsize error rc %d\n", rc);
628 * Get the value of the default_easize parameter.
630 * \see client_obd::cl_default_mds_easize
632 * \param[in] sbi superblock info for this filesystem
633 * \param[out] lmmsize pointer to storage location for value
635 * \retval 0 on success
636 * \retval negative negated errno on failure
638 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
643 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
644 KEY_DEFAULT_EASIZE, &size, lmmsize);
646 CERROR("Get default mdsize error rc %d\n", rc);
652 * Set the default_easize parameter to the given value.
654 * \see client_obd::cl_default_mds_easize
656 * \param[in] sbi superblock info for this filesystem
657 * \param[in] lmmsize the size to set
659 * \retval 0 on success
660 * \retval negative negated errno on failure
662 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
666 if (lmmsize < sizeof(struct lov_mds_md) ||
667 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
670 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
671 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
672 sizeof(int), &lmmsize, NULL);
677 static void client_common_put_super(struct super_block *sb)
679 struct ll_sb_info *sbi = ll_s2sbi(sb);
684 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
685 obd_disconnect(sbi->ll_dt_exp);
686 sbi->ll_dt_exp = NULL;
688 lprocfs_unregister_mountpoint(sbi);
690 obd_fid_fini(sbi->ll_md_exp->exp_obd);
691 obd_disconnect(sbi->ll_md_exp);
692 sbi->ll_md_exp = NULL;
697 void ll_kill_super(struct super_block *sb)
699 struct ll_sb_info *sbi;
703 if (!(sb->s_flags & MS_ACTIVE))
707 /* we need restore s_dev from changed for clustred NFS before put_super
708 * because new kernels have cached s_dev and change sb->s_dev in
709 * put_super not affected real removing devices */
711 sb->s_dev = sbi->ll_sdev_orig;
712 sbi->ll_umounting = 1;
714 /* wait running statahead threads to quit */
715 while (atomic_read(&sbi->ll_sa_running) > 0) {
716 set_current_state(TASK_UNINTERRUPTIBLE);
717 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
724 static inline int ll_set_opt(const char *opt, char *data, int fl)
726 if (strncmp(opt, data, strlen(opt)) != 0)
732 /* non-client-specific mount options are parsed in lmd_parse */
733 static int ll_options(char *options, int *flags)
736 char *s1 = options, *s2;
742 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
745 CDEBUG(D_SUPER, "next opt=%s\n", s1);
746 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
751 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
756 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
761 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
766 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
771 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
776 tmp = ll_set_opt("context", s1, 1);
779 tmp = ll_set_opt("fscontext", s1, 1);
782 tmp = ll_set_opt("defcontext", s1, 1);
785 tmp = ll_set_opt("rootcontext", s1, 1);
788 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
793 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
799 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
804 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
809 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
814 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
819 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
824 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
829 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
834 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
839 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
844 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
849 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
855 s2 = strchr(s1, ',');
863 void ll_lli_init(struct ll_inode_info *lli)
865 lli->lli_inode_magic = LLI_INODE_MAGIC;
867 spin_lock_init(&lli->lli_lock);
868 lli->lli_posix_acl = NULL;
869 /* Do not set lli_fid, it has been initialized already. */
870 fid_zero(&lli->lli_pfid);
871 lli->lli_mds_read_och = NULL;
872 lli->lli_mds_write_och = NULL;
873 lli->lli_mds_exec_och = NULL;
874 lli->lli_open_fd_read_count = 0;
875 lli->lli_open_fd_write_count = 0;
876 lli->lli_open_fd_exec_count = 0;
877 mutex_init(&lli->lli_och_mutex);
878 spin_lock_init(&lli->lli_agl_lock);
879 spin_lock_init(&lli->lli_layout_lock);
880 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
881 lli->lli_clob = NULL;
883 init_rwsem(&lli->lli_xattrs_list_rwsem);
884 mutex_init(&lli->lli_xattrs_enq_lock);
886 LASSERT(lli->lli_vfs_inode.i_mode != 0);
887 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
888 mutex_init(&lli->lli_readdir_mutex);
889 lli->lli_opendir_key = NULL;
891 spin_lock_init(&lli->lli_sa_lock);
892 lli->lli_opendir_pid = 0;
893 lli->lli_sa_enabled = 0;
894 lli->lli_def_stripe_offset = -1;
896 mutex_init(&lli->lli_size_mutex);
897 lli->lli_symlink_name = NULL;
898 init_rwsem(&lli->lli_trunc_sem);
899 range_lock_tree_init(&lli->lli_write_tree);
900 init_rwsem(&lli->lli_glimpse_sem);
901 lli->lli_glimpse_time = 0;
902 INIT_LIST_HEAD(&lli->lli_agl_list);
903 lli->lli_agl_index = 0;
904 lli->lli_async_rc = 0;
906 mutex_init(&lli->lli_layout_mutex);
907 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
910 static inline int ll_bdi_register(struct backing_dev_info *bdi)
912 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
914 bdi->name = "lustre";
915 return bdi_register(bdi, NULL, "lustre-%d",
916 atomic_inc_return(&ll_bdi_num));
919 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
921 struct lustre_profile *lprof = NULL;
922 struct lustre_sb_info *lsi = s2lsi(sb);
923 struct ll_sb_info *sbi;
924 char *dt = NULL, *md = NULL;
925 char *profilenm = get_profile_name(sb);
926 struct config_llog_instance *cfg;
927 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
928 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
932 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
938 try_module_get(THIS_MODULE);
940 /* client additional sb info */
941 lsi->lsi_llsbi = sbi = ll_init_sbi();
943 module_put(THIS_MODULE);
948 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
952 err = bdi_init(&lsi->lsi_bdi);
955 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
956 #ifdef HAVE_BDI_CAP_MAP_COPY
957 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
959 lsi->lsi_bdi.capabilities = 0;
961 err = ll_bdi_register(&lsi->lsi_bdi);
965 sb->s_bdi = &lsi->lsi_bdi;
966 #ifndef HAVE_DCACHE_LOCK
967 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
968 sb->s_d_op = &ll_d_ops;
971 /* Generate a string unique to this super, in case some joker tries
972 to mount the same fs at two mount points.
973 Use the address of the super itself.*/
974 cfg->cfg_instance = sb;
975 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
976 cfg->cfg_callback = class_config_llog_handler;
977 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
978 /* set up client obds */
979 err = lustre_process_log(sb, profilenm, cfg);
983 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
984 lprof = class_get_profile(profilenm);
986 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
987 " read from the MGS. Does that filesystem "
988 "exist?\n", profilenm);
989 GOTO(out_free, err = -EINVAL);
991 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
992 lprof->lp_md, lprof->lp_dt);
994 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
996 GOTO(out_free, err = -ENOMEM);
997 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
999 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1001 GOTO(out_free, err = -ENOMEM);
1002 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1004 /* connections, registrations, sb setup */
1005 err = client_common_fill_super(sb, md, dt, mnt);
1007 GOTO(out_free, err);
1009 sbi->ll_client_common_fill_super_succeeded = 1;
1013 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1015 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1017 class_put_profile(lprof);
1020 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1021 LCONSOLE_WARN("Mounted %s\n", profilenm);
1025 } /* ll_fill_super */
1027 void ll_put_super(struct super_block *sb)
1029 struct config_llog_instance cfg, params_cfg;
1030 struct obd_device *obd;
1031 struct lustre_sb_info *lsi = s2lsi(sb);
1032 struct ll_sb_info *sbi = ll_s2sbi(sb);
1033 char *profilenm = get_profile_name(sb);
1035 int next, force = 1, rc = 0;
1038 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1040 cfg.cfg_instance = sb;
1041 lustre_end_log(sb, profilenm, &cfg);
1043 params_cfg.cfg_instance = sb;
1044 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1046 if (sbi->ll_md_exp) {
1047 obd = class_exp2obd(sbi->ll_md_exp);
1049 force = obd->obd_force;
1052 /* Wait for unstable pages to be committed to stable storage */
1054 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1055 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1056 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1060 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1061 if (force == 0 && rc != -EINTR)
1062 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1065 /* We need to set force before the lov_disconnect in
1066 lustre_common_put_super, since l_d cleans up osc's as well. */
1069 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1071 obd->obd_force = force;
1075 if (sbi->ll_client_common_fill_super_succeeded) {
1076 /* Only if client_common_fill_super succeeded */
1077 client_common_put_super(sb);
1081 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1082 class_manual_cleanup(obd);
1085 if (sbi->ll_flags & LL_SBI_VERBOSE)
1086 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1089 class_del_profile(profilenm);
1091 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1092 bdi_destroy(&lsi->lsi_bdi);
1093 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1097 lsi->lsi_llsbi = NULL;
1099 lustre_common_put_super(sb);
1101 cl_env_cache_purge(~0);
1103 module_put(THIS_MODULE);
1106 } /* client_put_super */
1108 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1110 struct inode *inode = NULL;
1112 /* NOTE: we depend on atomic igrab() -bzzz */
1113 lock_res_and_lock(lock);
1114 if (lock->l_resource->lr_lvb_inode) {
1115 struct ll_inode_info * lli;
1116 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1117 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1118 inode = igrab(lock->l_resource->lr_lvb_inode);
1120 inode = lock->l_resource->lr_lvb_inode;
1121 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1122 D_WARNING, lock, "lr_lvb_inode %p is "
1123 "bogus: magic %08x",
1124 lock->l_resource->lr_lvb_inode,
1125 lli->lli_inode_magic);
1129 unlock_res_and_lock(lock);
1133 void ll_dir_clear_lsm_md(struct inode *inode)
1135 struct ll_inode_info *lli = ll_i2info(inode);
1137 LASSERT(S_ISDIR(inode->i_mode));
1139 if (lli->lli_lsm_md != NULL) {
1140 lmv_free_memmd(lli->lli_lsm_md);
1141 lli->lli_lsm_md = NULL;
1145 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1146 const struct lu_fid *fid,
1147 struct lustre_md *md)
1149 struct ll_sb_info *sbi = ll_s2sbi(sb);
1150 struct mdt_body *body = md->body;
1151 struct inode *inode;
1155 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1156 inode = iget_locked(sb, ino);
1157 if (inode == NULL) {
1158 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1159 ll_get_fsname(sb, NULL, 0), PFID(fid));
1160 RETURN(ERR_PTR(-ENOENT));
1163 if (inode->i_state & I_NEW) {
1164 struct ll_inode_info *lli = ll_i2info(inode);
1165 struct lmv_stripe_md *lsm = md->lmv;
1167 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1168 (body->mbo_mode & S_IFMT);
1169 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1172 LTIME_S(inode->i_mtime) = 0;
1173 LTIME_S(inode->i_atime) = 0;
1174 LTIME_S(inode->i_ctime) = 0;
1177 #ifdef HAVE_BACKING_DEV_INFO
1178 /* initializing backing dev info. */
1179 inode->i_mapping->backing_dev_info =
1180 &s2lsi(inode->i_sb)->lsi_bdi;
1182 inode->i_op = &ll_dir_inode_operations;
1183 inode->i_fop = &ll_dir_operations;
1184 lli->lli_fid = *fid;
1187 LASSERT(lsm != NULL);
1188 /* master object FID */
1189 lli->lli_pfid = body->mbo_fid1;
1190 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1191 lli, PFID(fid), PFID(&lli->lli_pfid));
1192 unlock_new_inode(inode);
1198 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1201 struct lmv_stripe_md *lsm = md->lmv;
1204 LASSERT(lsm != NULL);
1205 /* XXX sigh, this lsm_root initialization should be in
1206 * LMV layer, but it needs ll_iget right now, so we
1207 * put this here right now. */
1208 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1209 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1210 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1211 /* Unfortunately ll_iget will call ll_update_inode,
1212 * where the initialization of slave inode is slightly
1213 * different, so it reset lsm_md to NULL to avoid
1214 * initializing lsm for slave inode. */
1215 /* For migrating inode, master stripe and master object will
1216 * be same, so we only need assign this inode */
1217 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1218 lsm->lsm_md_oinfo[i].lmo_root = inode;
1220 lsm->lsm_md_oinfo[i].lmo_root =
1221 ll_iget_anon_dir(inode->i_sb, fid, md);
1223 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1224 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1226 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1234 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1235 const struct lmv_stripe_md *lsm_md2)
1237 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1238 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1239 lsm_md1->lsm_md_master_mdt_index ==
1240 lsm_md2->lsm_md_master_mdt_index &&
1241 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1242 lsm_md1->lsm_md_layout_version ==
1243 lsm_md2->lsm_md_layout_version &&
1244 strcmp(lsm_md1->lsm_md_pool_name,
1245 lsm_md2->lsm_md_pool_name) == 0;
1248 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1250 struct ll_inode_info *lli = ll_i2info(inode);
1251 struct lmv_stripe_md *lsm = md->lmv;
1255 LASSERT(S_ISDIR(inode->i_mode));
1256 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1257 PFID(ll_inode2fid(inode)));
1259 /* no striped information from request. */
1261 if (lli->lli_lsm_md == NULL) {
1263 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1264 LMV_HASH_FLAG_MIGRATION) {
1265 /* migration is done, the temporay MIGRATE layout has
1267 CDEBUG(D_INODE, DFID" finish migration.\n",
1268 PFID(ll_inode2fid(inode)));
1269 lmv_free_memmd(lli->lli_lsm_md);
1270 lli->lli_lsm_md = NULL;
1273 /* The lustre_md from req does not include stripeEA,
1274 * see ll_md_setattr */
1279 /* set the directory layout */
1280 if (lli->lli_lsm_md == NULL) {
1281 struct cl_attr *attr;
1283 rc = ll_init_lsm_md(inode, md);
1287 /* set md->lmv to NULL, so the following free lustre_md
1288 * will not free this lsm */
1290 lli->lli_lsm_md = lsm;
1292 OBD_ALLOC_PTR(attr);
1296 /* validate the lsm */
1297 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1298 ll_md_blocking_ast);
1304 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1305 md->body->mbo_nlink = attr->cat_nlink;
1306 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1307 md->body->mbo_size = attr->cat_size;
1308 if (md->body->mbo_valid & OBD_MD_FLATIME)
1309 md->body->mbo_atime = attr->cat_atime;
1310 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1311 md->body->mbo_ctime = attr->cat_ctime;
1312 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1313 md->body->mbo_mtime = attr->cat_mtime;
1317 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1318 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1322 /* Compare the old and new stripe information */
1323 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1324 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1327 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1328 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1329 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1330 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1331 inode, lsm, old_lsm,
1332 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1333 lsm->lsm_md_stripe_count,
1334 old_lsm->lsm_md_stripe_count,
1335 lsm->lsm_md_master_mdt_index,
1336 old_lsm->lsm_md_master_mdt_index,
1337 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1338 lsm->lsm_md_layout_version,
1339 old_lsm->lsm_md_layout_version,
1340 lsm->lsm_md_pool_name,
1341 old_lsm->lsm_md_pool_name);
1343 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1344 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1345 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1346 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1349 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1350 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1351 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1352 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1361 void ll_clear_inode(struct inode *inode)
1363 struct ll_inode_info *lli = ll_i2info(inode);
1364 struct ll_sb_info *sbi = ll_i2sbi(inode);
1367 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1368 PFID(ll_inode2fid(inode)), inode);
1370 if (S_ISDIR(inode->i_mode)) {
1371 /* these should have been cleared in ll_file_release */
1372 LASSERT(lli->lli_opendir_key == NULL);
1373 LASSERT(lli->lli_sai == NULL);
1374 LASSERT(lli->lli_opendir_pid == 0);
1377 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1379 LASSERT(!lli->lli_open_fd_write_count);
1380 LASSERT(!lli->lli_open_fd_read_count);
1381 LASSERT(!lli->lli_open_fd_exec_count);
1383 if (lli->lli_mds_write_och)
1384 ll_md_real_close(inode, FMODE_WRITE);
1385 if (lli->lli_mds_exec_och)
1386 ll_md_real_close(inode, FMODE_EXEC);
1387 if (lli->lli_mds_read_och)
1388 ll_md_real_close(inode, FMODE_READ);
1390 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1391 OBD_FREE(lli->lli_symlink_name,
1392 strlen(lli->lli_symlink_name) + 1);
1393 lli->lli_symlink_name = NULL;
1396 ll_xattr_cache_destroy(inode);
1398 #ifdef CONFIG_FS_POSIX_ACL
1399 if (lli->lli_posix_acl) {
1400 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1401 posix_acl_release(lli->lli_posix_acl);
1402 lli->lli_posix_acl = NULL;
1405 lli->lli_inode_magic = LLI_INODE_DEAD;
1407 if (S_ISDIR(inode->i_mode))
1408 ll_dir_clear_lsm_md(inode);
1409 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1410 LASSERT(list_empty(&lli->lli_agl_list));
1413 * XXX This has to be done before lsm is freed below, because
1414 * cl_object still uses inode lsm.
1416 cl_inode_fini(inode);
1421 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1423 struct lustre_md md;
1424 struct inode *inode = dentry->d_inode;
1425 struct ll_sb_info *sbi = ll_i2sbi(inode);
1426 struct ptlrpc_request *request = NULL;
1430 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1431 LUSTRE_OPC_ANY, NULL);
1432 if (IS_ERR(op_data))
1433 RETURN(PTR_ERR(op_data));
1435 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1437 ptlrpc_req_finished(request);
1438 if (rc == -ENOENT) {
1440 /* Unlinked special device node? Or just a race?
1441 * Pretend we done everything. */
1442 if (!S_ISREG(inode->i_mode) &&
1443 !S_ISDIR(inode->i_mode)) {
1444 ia_valid = op_data->op_attr.ia_valid;
1445 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1446 rc = simple_setattr(dentry, &op_data->op_attr);
1447 op_data->op_attr.ia_valid = ia_valid;
1449 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1450 CERROR("md_setattr fails: rc = %d\n", rc);
1455 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1456 sbi->ll_md_exp, &md);
1458 ptlrpc_req_finished(request);
1462 ia_valid = op_data->op_attr.ia_valid;
1463 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1464 * cache is not cleared yet. */
1465 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1466 if (S_ISREG(inode->i_mode))
1468 rc = simple_setattr(dentry, &op_data->op_attr);
1469 if (S_ISREG(inode->i_mode))
1470 inode_unlock(inode);
1471 op_data->op_attr.ia_valid = ia_valid;
1473 rc = ll_update_inode(inode, &md);
1474 ptlrpc_req_finished(request);
1479 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1480 * object(s) determine the file size and mtime. Otherwise, the MDS will
1481 * keep these values until such a time that objects are allocated for it.
1482 * We do the MDS operations first, as it is checking permissions for us.
1483 * We don't to the MDS RPC if there is nothing that we want to store there,
1484 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1485 * going to do an RPC anyways.
1487 * If we are doing a truncate, we will send the mtime and ctime updates
1488 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1489 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1492 * In case of HSMimport, we only set attr on MDS.
1494 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1496 struct inode *inode = dentry->d_inode;
1497 struct ll_inode_info *lli = ll_i2info(inode);
1498 struct md_op_data *op_data = NULL;
1502 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1503 "valid %x, hsm_import %d\n",
1504 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1505 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1508 if (attr->ia_valid & ATTR_SIZE) {
1509 /* Check new size against VFS/VM file size limit and rlimit */
1510 rc = inode_newsize_ok(inode, attr->ia_size);
1514 /* The maximum Lustre file size is variable, based on the
1515 * OST maximum object size and number of stripes. This
1516 * needs another check in addition to the VFS check above. */
1517 if (attr->ia_size > ll_file_maxbytes(inode)) {
1518 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1519 PFID(&lli->lli_fid), attr->ia_size,
1520 ll_file_maxbytes(inode));
1524 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1527 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1528 if (attr->ia_valid & TIMES_SET_FLAGS) {
1529 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1530 !cfs_capable(CFS_CAP_FOWNER))
1534 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1535 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1536 (attr->ia_valid & ATTR_CTIME)) {
1537 attr->ia_ctime = CURRENT_TIME;
1538 attr->ia_valid |= ATTR_CTIME_SET;
1540 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1541 (attr->ia_valid & ATTR_ATIME)) {
1542 attr->ia_atime = CURRENT_TIME;
1543 attr->ia_valid |= ATTR_ATIME_SET;
1545 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1546 (attr->ia_valid & ATTR_MTIME)) {
1547 attr->ia_mtime = CURRENT_TIME;
1548 attr->ia_valid |= ATTR_MTIME_SET;
1551 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1552 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1553 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1554 (s64)ktime_get_real_seconds());
1556 if (S_ISREG(inode->i_mode)) {
1557 if (attr->ia_valid & ATTR_SIZE)
1558 inode_dio_write_done(inode);
1559 inode_unlock(inode);
1562 /* We always do an MDS RPC, even if we're only changing the size;
1563 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1565 OBD_ALLOC_PTR(op_data);
1566 if (op_data == NULL)
1567 GOTO(out, rc = -ENOMEM);
1569 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1570 /* If we are changing file size, file content is
1571 * modified, flag it. */
1572 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1573 op_data->op_bias |= MDS_DATA_MODIFIED;
1574 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1577 op_data->op_attr = *attr;
1579 rc = ll_md_setattr(dentry, op_data);
1583 if (!S_ISREG(inode->i_mode) || hsm_import)
1586 if (attr->ia_valid & (ATTR_SIZE |
1587 ATTR_ATIME | ATTR_ATIME_SET |
1588 ATTR_MTIME | ATTR_MTIME_SET |
1589 ATTR_CTIME | ATTR_CTIME_SET)) {
1590 /* For truncate and utimes sending attributes to OSTs, setting
1591 * mtime/atime to the past will be performed under PW [0:EOF]
1592 * extent lock (new_size:EOF for truncate). It may seem
1593 * excessive to send mtime/atime updates to OSTs when not
1594 * setting times to past, but it is necessary due to possible
1595 * time de-synchronization between MDT inode and OST objects */
1596 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1599 /* If the file was restored, it needs to set dirty flag.
1601 * We've already sent MDS_DATA_MODIFIED flag in
1602 * ll_md_setattr() for truncate. However, the MDT refuses to
1603 * set the HS_DIRTY flag on released files, so we have to set
1604 * it again if the file has been restored. Please check how
1605 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1607 * Please notice that if the file is not released, the previous
1608 * MDS_DATA_MODIFIED has taken effect and usually
1609 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1610 * This way we can save an RPC for common open + trunc
1612 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1613 struct hsm_state_set hss = {
1614 .hss_valid = HSS_SETMASK,
1615 .hss_setmask = HS_DIRTY,
1619 rc2 = ll_hsm_state_set(inode, &hss);
1620 /* truncate and write can happen at the same time, so that
1621 * the file can be set modified even though the file is not
1622 * restored from released state, and ll_hsm_state_set() is
1623 * not applicable for the file, and rc2 < 0 is normal in this
1626 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1627 PFID(ll_inode2fid(inode)), rc2);
1632 if (op_data != NULL)
1633 ll_finish_md_op_data(op_data);
1635 if (S_ISREG(inode->i_mode)) {
1637 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1638 inode_dio_wait(inode);
1641 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1642 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1647 int ll_setattr(struct dentry *de, struct iattr *attr)
1649 int mode = de->d_inode->i_mode;
1651 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1652 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1653 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1655 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1656 (ATTR_SIZE|ATTR_MODE)) &&
1657 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1658 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1659 !(attr->ia_mode & S_ISGID))))
1660 attr->ia_valid |= ATTR_FORCE;
1662 if ((attr->ia_valid & ATTR_MODE) &&
1664 !(attr->ia_mode & S_ISUID) &&
1665 !(attr->ia_valid & ATTR_KILL_SUID))
1666 attr->ia_valid |= ATTR_KILL_SUID;
1668 if ((attr->ia_valid & ATTR_MODE) &&
1669 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1670 !(attr->ia_mode & S_ISGID) &&
1671 !(attr->ia_valid & ATTR_KILL_SGID))
1672 attr->ia_valid |= ATTR_KILL_SGID;
1674 /* avoid polluted from ATTR_TIMES_SET,
1675 * projid is not expected to be set here */
1676 attr->ia_valid &= ~MDS_ATTR_PROJID;
1678 return ll_setattr_raw(de, attr, false);
1681 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1682 __u64 max_age, __u32 flags)
1684 struct ll_sb_info *sbi = ll_s2sbi(sb);
1685 struct obd_statfs obd_osfs;
1689 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1691 CERROR("md_statfs fails: rc = %d\n", rc);
1695 osfs->os_type = sb->s_magic;
1697 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1698 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1700 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1701 flags |= OBD_STATFS_NODELAY;
1703 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1705 CERROR("obd_statfs fails: rc = %d\n", rc);
1709 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1710 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1713 osfs->os_bsize = obd_osfs.os_bsize;
1714 osfs->os_blocks = obd_osfs.os_blocks;
1715 osfs->os_bfree = obd_osfs.os_bfree;
1716 osfs->os_bavail = obd_osfs.os_bavail;
1718 /* If we don't have as many objects free on the OST as inodes
1719 * on the MDS, we reduce the total number of inodes to
1720 * compensate, so that the "inodes in use" number is correct.
1722 if (obd_osfs.os_ffree < osfs->os_ffree) {
1723 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1725 osfs->os_ffree = obd_osfs.os_ffree;
1730 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1732 struct super_block *sb = de->d_sb;
1733 struct obd_statfs osfs;
1734 __u64 fsid = huge_encode_dev(sb->s_dev);
1737 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1738 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1740 /* Some amount of caching on the client is allowed */
1741 rc = ll_statfs_internal(sb, &osfs,
1742 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1747 statfs_unpack(sfs, &osfs);
1749 /* We need to downshift for all 32-bit kernels, because we can't
1750 * tell if the kernel is being called via sys_statfs64() or not.
1751 * Stop before overflowing f_bsize - in which case it is better
1752 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1753 if (sizeof(long) < 8) {
1754 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1757 osfs.os_blocks >>= 1;
1758 osfs.os_bfree >>= 1;
1759 osfs.os_bavail >>= 1;
1763 sfs->f_blocks = osfs.os_blocks;
1764 sfs->f_bfree = osfs.os_bfree;
1765 sfs->f_bavail = osfs.os_bavail;
1766 sfs->f_fsid.val[0] = (__u32)fsid;
1767 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1771 void ll_inode_size_lock(struct inode *inode)
1773 struct ll_inode_info *lli;
1775 LASSERT(!S_ISDIR(inode->i_mode));
1777 lli = ll_i2info(inode);
1778 mutex_lock(&lli->lli_size_mutex);
1781 void ll_inode_size_unlock(struct inode *inode)
1783 struct ll_inode_info *lli;
1785 lli = ll_i2info(inode);
1786 mutex_unlock(&lli->lli_size_mutex);
1789 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1791 struct ll_inode_info *lli = ll_i2info(inode);
1792 struct mdt_body *body = md->body;
1793 struct ll_sb_info *sbi = ll_i2sbi(inode);
1795 if (body->mbo_valid & OBD_MD_FLEASIZE)
1796 cl_file_inode_init(inode, md);
1798 if (S_ISDIR(inode->i_mode)) {
1801 rc = ll_update_lsm_md(inode, md);
1806 #ifdef CONFIG_FS_POSIX_ACL
1807 if (body->mbo_valid & OBD_MD_FLACL) {
1808 spin_lock(&lli->lli_lock);
1809 if (lli->lli_posix_acl)
1810 posix_acl_release(lli->lli_posix_acl);
1811 lli->lli_posix_acl = md->posix_acl;
1812 spin_unlock(&lli->lli_lock);
1815 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1816 sbi->ll_flags & LL_SBI_32BIT_API);
1817 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1819 if (body->mbo_valid & OBD_MD_FLATIME) {
1820 if (body->mbo_atime > LTIME_S(inode->i_atime))
1821 LTIME_S(inode->i_atime) = body->mbo_atime;
1822 lli->lli_atime = body->mbo_atime;
1825 if (body->mbo_valid & OBD_MD_FLMTIME) {
1826 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1827 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1828 "to %llu\n", inode->i_ino,
1829 LTIME_S(inode->i_mtime), body->mbo_mtime);
1830 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1832 lli->lli_mtime = body->mbo_mtime;
1835 if (body->mbo_valid & OBD_MD_FLCTIME) {
1836 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1837 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1838 lli->lli_ctime = body->mbo_ctime;
1841 if (body->mbo_valid & OBD_MD_FLMODE)
1842 inode->i_mode = (inode->i_mode & S_IFMT) |
1843 (body->mbo_mode & ~S_IFMT);
1845 if (body->mbo_valid & OBD_MD_FLTYPE)
1846 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1847 (body->mbo_mode & S_IFMT);
1849 LASSERT(inode->i_mode != 0);
1850 if (S_ISREG(inode->i_mode))
1851 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1852 LL_MAX_BLKSIZE_BITS);
1854 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1856 if (body->mbo_valid & OBD_MD_FLUID)
1857 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1858 if (body->mbo_valid & OBD_MD_FLGID)
1859 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1860 if (body->mbo_valid & OBD_MD_FLPROJID)
1861 lli->lli_projid = body->mbo_projid;
1862 if (body->mbo_valid & OBD_MD_FLFLAGS)
1863 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1864 if (body->mbo_valid & OBD_MD_FLNLINK)
1865 set_nlink(inode, body->mbo_nlink);
1866 if (body->mbo_valid & OBD_MD_FLRDEV)
1867 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1869 if (body->mbo_valid & OBD_MD_FLID) {
1870 /* FID shouldn't be changed! */
1871 if (fid_is_sane(&lli->lli_fid)) {
1872 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1873 "Trying to change FID "DFID
1874 " to the "DFID", inode "DFID"(%p)\n",
1875 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1876 PFID(ll_inode2fid(inode)), inode);
1878 lli->lli_fid = body->mbo_fid1;
1882 LASSERT(fid_seq(&lli->lli_fid) != 0);
1884 if (body->mbo_valid & OBD_MD_FLSIZE) {
1885 i_size_write(inode, body->mbo_size);
1887 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1888 PFID(ll_inode2fid(inode)),
1889 (unsigned long long)body->mbo_size);
1891 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1892 inode->i_blocks = body->mbo_blocks;
1895 if (body->mbo_valid & OBD_MD_TSTATE) {
1896 /* Set LLIF_FILE_RESTORING if restore ongoing and
1897 * clear it when done to ensure to start again
1898 * glimpsing updated attrs
1900 if (body->mbo_t_state & MS_RESTORE)
1901 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1903 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1909 int ll_read_inode2(struct inode *inode, void *opaque)
1911 struct lustre_md *md = opaque;
1912 struct ll_inode_info *lli = ll_i2info(inode);
1916 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1917 PFID(&lli->lli_fid), inode);
1919 /* Core attributes from the MDS first. This is a new inode, and
1920 * the VFS doesn't zero times in the core inode so we have to do
1921 * it ourselves. They will be overwritten by either MDS or OST
1922 * attributes - we just need to make sure they aren't newer. */
1923 LTIME_S(inode->i_mtime) = 0;
1924 LTIME_S(inode->i_atime) = 0;
1925 LTIME_S(inode->i_ctime) = 0;
1927 rc = ll_update_inode(inode, md);
1931 /* OIDEBUG(inode); */
1933 #ifdef HAVE_BACKING_DEV_INFO
1934 /* initializing backing dev info. */
1935 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1937 if (S_ISREG(inode->i_mode)) {
1938 struct ll_sb_info *sbi = ll_i2sbi(inode);
1939 inode->i_op = &ll_file_inode_operations;
1940 inode->i_fop = sbi->ll_fop;
1941 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1943 } else if (S_ISDIR(inode->i_mode)) {
1944 inode->i_op = &ll_dir_inode_operations;
1945 inode->i_fop = &ll_dir_operations;
1947 } else if (S_ISLNK(inode->i_mode)) {
1948 inode->i_op = &ll_fast_symlink_inode_operations;
1951 inode->i_op = &ll_special_inode_operations;
1953 init_special_inode(inode, inode->i_mode,
1962 void ll_delete_inode(struct inode *inode)
1964 struct ll_inode_info *lli = ll_i2info(inode);
1967 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1968 /* It is last chance to write out dirty pages,
1969 * otherwise we may lose data while umount */
1970 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1972 truncate_inode_pages_final(&inode->i_data);
1974 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
1975 "see https://jira.hpdd.intel.com/browse/LU-118\n",
1976 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
1978 #ifdef HAVE_SBOPS_EVICT_INODE
1979 ll_clear_inode(inode);
1986 int ll_iocontrol(struct inode *inode, struct file *file,
1987 unsigned int cmd, unsigned long arg)
1989 struct ll_sb_info *sbi = ll_i2sbi(inode);
1990 struct ptlrpc_request *req = NULL;
1995 case FSFILT_IOC_GETFLAGS: {
1996 struct mdt_body *body;
1997 struct md_op_data *op_data;
1999 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2000 0, 0, LUSTRE_OPC_ANY,
2002 if (IS_ERR(op_data))
2003 RETURN(PTR_ERR(op_data));
2005 op_data->op_valid = OBD_MD_FLFLAGS;
2006 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2007 ll_finish_md_op_data(op_data);
2009 CERROR("%s: failure inode "DFID": rc = %d\n",
2010 sbi->ll_md_exp->exp_obd->obd_name,
2011 PFID(ll_inode2fid(inode)), rc);
2015 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2017 flags = body->mbo_flags;
2019 ptlrpc_req_finished(req);
2021 RETURN(put_user(flags, (int __user *)arg));
2023 case FSFILT_IOC_SETFLAGS: {
2025 struct md_op_data *op_data;
2026 struct cl_object *obj;
2028 if (get_user(flags, (int __user *)arg))
2031 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2032 LUSTRE_OPC_ANY, NULL);
2033 if (IS_ERR(op_data))
2034 RETURN(PTR_ERR(op_data));
2036 op_data->op_attr_flags = flags;
2037 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2038 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2039 ll_finish_md_op_data(op_data);
2040 ptlrpc_req_finished(req);
2044 inode->i_flags = ll_ext_to_inode_flags(flags);
2046 obj = ll_i2info(inode)->lli_clob;
2050 OBD_ALLOC_PTR(attr);
2054 attr->ia_valid = ATTR_ATTR_FLAG;
2055 rc = cl_setattr_ost(obj, attr, flags);
2067 int ll_flush_ctx(struct inode *inode)
2069 struct ll_sb_info *sbi = ll_i2sbi(inode);
2071 CDEBUG(D_SEC, "flush context for user %d\n",
2072 from_kuid(&init_user_ns, current_uid()));
2074 obd_set_info_async(NULL, sbi->ll_md_exp,
2075 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2077 obd_set_info_async(NULL, sbi->ll_dt_exp,
2078 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2083 /* umount -f client means force down, don't save state */
2084 void ll_umount_begin(struct super_block *sb)
2086 struct ll_sb_info *sbi = ll_s2sbi(sb);
2087 struct obd_device *obd;
2088 struct obd_ioctl_data *ioc_data;
2089 struct l_wait_info lwi;
2090 wait_queue_head_t waitq;
2093 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2094 sb->s_count, atomic_read(&sb->s_active));
2096 obd = class_exp2obd(sbi->ll_md_exp);
2098 CERROR("Invalid MDC connection handle %#llx\n",
2099 sbi->ll_md_exp->exp_handle.h_cookie);
2105 obd = class_exp2obd(sbi->ll_dt_exp);
2107 CERROR("Invalid LOV connection handle %#llx\n",
2108 sbi->ll_dt_exp->exp_handle.h_cookie);
2114 OBD_ALLOC_PTR(ioc_data);
2116 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2117 sizeof *ioc_data, ioc_data, NULL);
2119 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2120 sizeof *ioc_data, ioc_data, NULL);
2122 OBD_FREE_PTR(ioc_data);
2125 /* Really, we'd like to wait until there are no requests outstanding,
2126 * and then continue. For now, we just periodically checking for vfs
2127 * to decrement mnt_cnt and hope to finish it within 10sec.
2129 init_waitqueue_head(&waitq);
2130 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2131 cfs_time_seconds(1), NULL, NULL);
2132 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2137 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2139 struct ll_sb_info *sbi = ll_s2sbi(sb);
2140 char *profilenm = get_profile_name(sb);
2144 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2145 read_only = *flags & MS_RDONLY;
2146 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2147 sizeof(KEY_READ_ONLY),
2148 KEY_READ_ONLY, sizeof(read_only),
2151 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2152 profilenm, read_only ?
2153 "read-only" : "read-write", err);
2158 sb->s_flags |= MS_RDONLY;
2160 sb->s_flags &= ~MS_RDONLY;
2162 if (sbi->ll_flags & LL_SBI_VERBOSE)
2163 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2164 read_only ? "read-only" : "read-write");
2170 * Cleanup the open handle that is cached on MDT-side.
2172 * For open case, the client side open handling thread may hit error
2173 * after the MDT grant the open. Under such case, the client should
2174 * send close RPC to the MDT as cleanup; otherwise, the open handle
2175 * on the MDT will be leaked there until the client umount or evicted.
2177 * In further, if someone unlinked the file, because the open handle
2178 * holds the reference on such file/object, then it will block the
2179 * subsequent threads that want to locate such object via FID.
2181 * \param[in] sb super block for this file-system
2182 * \param[in] open_req pointer to the original open request
2184 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2186 struct mdt_body *body;
2187 struct md_op_data *op_data;
2188 struct ptlrpc_request *close_req = NULL;
2189 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2192 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2193 OBD_ALLOC_PTR(op_data);
2194 if (op_data == NULL) {
2195 CWARN("%s: cannot allocate op_data to release open handle for "
2197 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2202 op_data->op_fid1 = body->mbo_fid1;
2203 op_data->op_handle = body->mbo_handle;
2204 op_data->op_mod_time = ktime_get_real_seconds();
2205 md_close(exp, op_data, NULL, &close_req);
2206 ptlrpc_req_finished(close_req);
2207 ll_finish_md_op_data(op_data);
2212 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2213 struct super_block *sb, struct lookup_intent *it)
2215 struct ll_sb_info *sbi = NULL;
2216 struct lustre_md md = { NULL };
2220 LASSERT(*inode || sb);
2221 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2222 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2223 sbi->ll_md_exp, &md);
2228 rc = ll_update_inode(*inode, &md);
2232 LASSERT(sb != NULL);
2235 * At this point server returns to client's same fid as client
2236 * generated for creating. So using ->fid1 is okay here.
2238 if (!fid_is_sane(&md.body->mbo_fid1)) {
2239 CERROR("%s: Fid is insane "DFID"\n",
2240 ll_get_fsname(sb, NULL, 0),
2241 PFID(&md.body->mbo_fid1));
2242 GOTO(out, rc = -EINVAL);
2245 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2246 sbi->ll_flags & LL_SBI_32BIT_API),
2248 if (IS_ERR(*inode)) {
2249 #ifdef CONFIG_FS_POSIX_ACL
2251 posix_acl_release(md.posix_acl);
2252 md.posix_acl = NULL;
2255 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2257 CERROR("new_inode -fatal: rc %d\n", rc);
2262 /* Handling piggyback layout lock.
2263 * Layout lock can be piggybacked by getattr and open request.
2264 * The lsm can be applied to inode only if it comes with a layout lock
2265 * otherwise correct layout may be overwritten, for example:
2266 * 1. proc1: mdt returns a lsm but not granting layout
2267 * 2. layout was changed by another client
2268 * 3. proc2: refresh layout and layout lock granted
2269 * 4. proc1: to apply a stale layout */
2270 if (it != NULL && it->it_lock_mode != 0) {
2271 struct lustre_handle lockh;
2272 struct ldlm_lock *lock;
2274 lockh.cookie = it->it_lock_handle;
2275 lock = ldlm_handle2lock(&lockh);
2276 LASSERT(lock != NULL);
2277 if (ldlm_has_layout(lock)) {
2278 struct cl_object_conf conf;
2280 memset(&conf, 0, sizeof(conf));
2281 conf.coc_opc = OBJECT_CONF_SET;
2282 conf.coc_inode = *inode;
2283 conf.coc_lock = lock;
2284 conf.u.coc_layout = md.layout;
2285 (void)ll_layout_conf(*inode, &conf);
2287 LDLM_LOCK_PUT(lock);
2293 md_free_lustre_md(sbi->ll_md_exp, &md);
2296 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2297 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2302 int ll_obd_statfs(struct inode *inode, void __user *arg)
2304 struct ll_sb_info *sbi = NULL;
2305 struct obd_export *exp;
2307 struct obd_ioctl_data *data = NULL;
2311 if (!inode || !(sbi = ll_i2sbi(inode)))
2312 GOTO(out_statfs, rc = -EINVAL);
2314 rc = obd_ioctl_getdata(&buf, &len, arg);
2316 GOTO(out_statfs, rc);
2319 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2320 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2321 GOTO(out_statfs, rc = -EINVAL);
2323 if (data->ioc_inllen1 != sizeof(__u32) ||
2324 data->ioc_inllen2 != sizeof(__u32) ||
2325 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2326 data->ioc_plen2 != sizeof(struct obd_uuid))
2327 GOTO(out_statfs, rc = -EINVAL);
2329 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2330 if (type & LL_STATFS_LMV)
2331 exp = sbi->ll_md_exp;
2332 else if (type & LL_STATFS_LOV)
2333 exp = sbi->ll_dt_exp;
2335 GOTO(out_statfs, rc = -ENODEV);
2337 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2339 GOTO(out_statfs, rc);
2341 OBD_FREE_LARGE(buf, len);
2345 int ll_process_config(struct lustre_cfg *lcfg)
2347 struct super_block *sb;
2352 /* The instance name contains the sb: lustre-client-aacfe000 */
2353 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2354 if (!ptr || !*(++ptr))
2356 if (sscanf(ptr, "%lx", &x) != 1)
2358 sb = (struct super_block *)x;
2359 /* This better be a real Lustre superblock! */
2360 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2362 /* Note we have not called client_common_fill_super yet, so
2363 proc fns must be able to handle that! */
2364 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2371 /* this function prepares md_op_data hint for passing it down to MD stack. */
2372 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2373 struct inode *i1, struct inode *i2,
2374 const char *name, size_t namelen,
2375 __u32 mode, __u32 opc, void *data)
2377 LASSERT(i1 != NULL);
2380 /* Do not reuse namelen for something else. */
2382 return ERR_PTR(-EINVAL);
2384 if (namelen > ll_i2sbi(i1)->ll_namelen)
2385 return ERR_PTR(-ENAMETOOLONG);
2387 if (!lu_name_is_valid_2(name, namelen))
2388 return ERR_PTR(-EINVAL);
2391 if (op_data == NULL)
2392 OBD_ALLOC_PTR(op_data);
2394 if (op_data == NULL)
2395 return ERR_PTR(-ENOMEM);
2397 ll_i2gids(op_data->op_suppgids, i1, i2);
2398 op_data->op_fid1 = *ll_inode2fid(i1);
2399 op_data->op_default_stripe_offset = -1;
2400 if (S_ISDIR(i1->i_mode)) {
2401 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2402 if (opc == LUSTRE_OPC_MKDIR)
2403 op_data->op_default_stripe_offset =
2404 ll_i2info(i1)->lli_def_stripe_offset;
2408 op_data->op_fid2 = *ll_inode2fid(i2);
2409 if (S_ISDIR(i2->i_mode))
2410 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2412 fid_zero(&op_data->op_fid2);
2415 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2416 op_data->op_cli_flags |= CLI_HASH64;
2418 if (ll_need_32bit_api(ll_i2sbi(i1)))
2419 op_data->op_cli_flags |= CLI_API32;
2421 op_data->op_name = name;
2422 op_data->op_namelen = namelen;
2423 op_data->op_mode = mode;
2424 op_data->op_mod_time = cfs_time_current_sec();
2425 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2426 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2427 op_data->op_cap = cfs_curproc_cap_pack();
2428 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2429 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2430 op_data->op_bias |= MDS_CREATE_VOLATILE;
2432 op_data->op_mds = 0;
2434 op_data->op_data = data;
2439 void ll_finish_md_op_data(struct md_op_data *op_data)
2441 security_release_secctx(op_data->op_file_secctx,
2442 op_data->op_file_secctx_size);
2443 OBD_FREE_PTR(op_data);
2446 #ifdef HAVE_SUPEROPS_USE_DENTRY
2447 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2449 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2452 struct ll_sb_info *sbi;
2454 #ifdef HAVE_SUPEROPS_USE_DENTRY
2455 LASSERT((seq != NULL) && (dentry != NULL));
2456 sbi = ll_s2sbi(dentry->d_sb);
2458 LASSERT((seq != NULL) && (vfs != NULL));
2459 sbi = ll_s2sbi(vfs->mnt_sb);
2462 if (sbi->ll_flags & LL_SBI_NOLCK)
2463 seq_puts(seq, ",nolock");
2465 if (sbi->ll_flags & LL_SBI_FLOCK)
2466 seq_puts(seq, ",flock");
2468 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2469 seq_puts(seq, ",localflock");
2471 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2472 seq_puts(seq, ",user_xattr");
2474 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2475 seq_puts(seq, ",lazystatfs");
2477 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2478 seq_puts(seq, ",user_fid2path");
2480 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2481 seq_puts(seq, ",always_ping");
2487 * Get obd name by cmd, and copy out to user space
2489 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2491 struct ll_sb_info *sbi = ll_i2sbi(inode);
2492 struct obd_device *obd;
2495 if (cmd == OBD_IOC_GETDTNAME)
2496 obd = class_exp2obd(sbi->ll_dt_exp);
2497 else if (cmd == OBD_IOC_GETMDNAME)
2498 obd = class_exp2obd(sbi->ll_md_exp);
2505 if (copy_to_user((void __user *)arg, obd->obd_name,
2506 strlen(obd->obd_name) + 1))
2513 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2514 * fsname will be returned in this buffer; otherwise, a static buffer will be
2515 * used to store the fsname and returned to caller.
2517 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2519 static char fsname_static[MTI_NAME_MAXLEN];
2520 struct lustre_sb_info *lsi = s2lsi(sb);
2525 /* this means the caller wants to use static buffer
2526 * and it doesn't care about race. Usually this is
2527 * in error reporting path */
2528 buf = fsname_static;
2529 buflen = sizeof(fsname_static);
2532 len = strlen(lsi->lsi_lmd->lmd_profile);
2533 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2534 if (ptr && (strcmp(ptr, "-client") == 0))
2537 if (unlikely(len >= buflen))
2539 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2545 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2552 p.mnt = current->fs->root.mnt;
2554 path = d_path(&p, buf, bufsize);
2559 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2561 char *buf, *path = NULL;
2562 struct dentry *dentry = NULL;
2563 struct inode *inode = page->mapping->host;
2565 /* this can be called inside spin lock so use GFP_ATOMIC. */
2566 buf = (char *)__get_free_page(GFP_ATOMIC);
2568 dentry = d_find_alias(page->mapping->host);
2570 path = ll_d_path(dentry, buf, PAGE_SIZE);
2574 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2575 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2576 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2577 PFID(ll_inode2fid(inode)),
2578 (path && !IS_ERR(path)) ? path : "", ioret);
2584 free_page((unsigned long)buf);
2587 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2588 struct lov_user_md **kbuf)
2590 struct lov_user_md lum;
2594 if (copy_from_user(&lum, md, sizeof(lum)))
2597 lum_size = ll_lov_user_md_size(&lum);
2601 OBD_ALLOC(*kbuf, lum_size);
2605 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2606 OBD_FREE(*kbuf, lum_size);
2614 * Compute llite root squash state after a change of root squash
2615 * configuration setting or add/remove of a lnet nid
2617 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2619 struct root_squash_info *squash = &sbi->ll_squash;
2622 struct lnet_process_id id;
2624 /* Update norootsquash flag */
2625 down_write(&squash->rsi_sem);
2626 if (list_empty(&squash->rsi_nosquash_nids))
2627 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2629 /* Do not apply root squash as soon as one of our NIDs is
2630 * in the nosquash_nids list */
2633 while (LNetGetId(i++, &id) != -ENOENT) {
2634 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2636 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2642 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2644 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2646 up_write(&squash->rsi_sem);
2650 * Parse linkea content to extract information about a given hardlink
2652 * \param[in] ldata - Initialized linkea data
2653 * \param[in] linkno - Link identifier
2654 * \param[out] parent_fid - The entry's parent FID
2655 * \param[out] ln - Entry name destination buffer
2657 * \retval 0 on success
2658 * \retval Appropriate negative error code on failure
2660 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2661 struct lu_fid *parent_fid, struct lu_name *ln)
2667 rc = linkea_init_with_rec(ldata);
2671 if (linkno >= ldata->ld_leh->leh_reccount)
2672 /* beyond last link */
2675 linkea_first_entry(ldata);
2676 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2677 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2682 linkea_next_entry(ldata);
2692 * Get parent FID and name of an identified link. Operation is performed for
2693 * a given link number, letting the caller iterate over linkno to list one or
2694 * all links of an entry.
2696 * \param[in] file - File descriptor against which to perform the operation
2697 * \param[in,out] arg - User-filled structure containing the linkno to operate
2698 * on and the available size. It is eventually filled with
2699 * the requested information or left untouched on error
2701 * \retval - 0 on success
2702 * \retval - Appropriate negative error code on failure
2704 int ll_getparent(struct file *file, struct getparent __user *arg)
2706 struct dentry *dentry = file_dentry(file);
2707 struct inode *inode = file_inode(file);
2708 struct linkea_data *ldata;
2709 struct lu_buf buf = LU_BUF_NULL;
2711 struct lu_fid parent_fid;
2718 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2719 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2722 if (get_user(name_size, &arg->gp_name_size))
2725 if (get_user(linkno, &arg->gp_linkno))
2728 if (name_size > PATH_MAX)
2731 OBD_ALLOC(ldata, sizeof(*ldata));
2735 rc = linkea_data_new(ldata, &buf);
2737 GOTO(ldata_free, rc);
2739 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2743 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2747 if (ln.ln_namelen >= name_size)
2748 GOTO(lb_free, rc = -EOVERFLOW);
2750 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2751 GOTO(lb_free, rc = -EFAULT);
2753 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2754 GOTO(lb_free, rc = -EFAULT);
2756 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2757 GOTO(lb_free, rc = -EFAULT);
2762 OBD_FREE(ldata, sizeof(*ldata));