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, 2015, 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 <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 <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 |
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;
230 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
231 /* flag mdc connection as lightweight, only used for test
232 * purpose, use with care */
233 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
235 data->ocd_ibits_known = MDS_INODELOCK_FULL;
236 data->ocd_version = LUSTRE_VERSION_CODE;
238 if (sb->s_flags & MS_RDONLY)
239 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
240 if (sbi->ll_flags & LL_SBI_USER_XATTR)
241 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
243 if (sbi->ll_flags & LL_SBI_FLOCK)
244 sbi->ll_fop = &ll_file_operations_flock;
245 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
246 sbi->ll_fop = &ll_file_operations;
248 sbi->ll_fop = &ll_file_operations_noflock;
250 /* always ping even if server suppress_pings */
251 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
252 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
254 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
255 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
256 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
258 data->ocd_brw_size = MD_MAX_BRW_SIZE;
260 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
262 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
263 "recovery, of which this client is not a "
264 "part. Please wait for recovery to complete,"
265 " abort, or time out.\n", md);
268 CERROR("cannot connect to %s: rc = %d\n", md, err);
272 sbi->ll_md_exp->exp_connect_data = *data;
274 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
275 LUSTRE_SEQ_METADATA);
277 CERROR("%s: Can't init metadata layer FID infrastructure, "
278 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
282 /* For mount, we only need fs info from MDT0, and also in DNE, it
283 * can make sure the client can be mounted as long as MDT0 is
285 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
286 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
287 OBD_STATFS_FOR_MDT0);
289 GOTO(out_md_fid, err);
291 /* This needs to be after statfs to ensure connect has finished.
292 * Note that "data" does NOT contain the valid connect reply.
293 * If connecting to a 1.8 server there will be no LMV device, so
294 * we can access the MDC export directly and exp_connect_flags will
295 * be non-zero, but if accessing an upgraded 2.1 server it will
296 * have the correct flags filled in.
297 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
298 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
299 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
300 valid != CLIENT_CONNECT_MDT_REQD) {
303 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
304 obd_connect_flags2str(buf, PAGE_SIZE,
305 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
306 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
307 "feature(s) needed for correct operation "
308 "of this client (%s). Please upgrade "
309 "server or downgrade client.\n",
310 sbi->ll_md_exp->exp_obd->obd_name, buf);
311 OBD_FREE(buf, PAGE_SIZE);
312 GOTO(out_md_fid, err = -EPROTO);
315 size = sizeof(*data);
316 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
317 KEY_CONN_DATA, &size, data);
319 CERROR("%s: Get connect data failed: rc = %d\n",
320 sbi->ll_md_exp->exp_obd->obd_name, err);
321 GOTO(out_md_fid, err);
324 LASSERT(osfs->os_bsize);
325 sb->s_blocksize = osfs->os_bsize;
326 sb->s_blocksize_bits = log2(osfs->os_bsize);
327 sb->s_magic = LL_SUPER_MAGIC;
328 sb->s_maxbytes = MAX_LFS_FILESIZE;
329 sbi->ll_namelen = osfs->os_namelen;
330 sbi->ll_mnt.mnt = current->fs->root.mnt;
332 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
333 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
334 LCONSOLE_INFO("Disabling user_xattr feature because "
335 "it is not supported on the server\n");
336 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
339 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
341 sb->s_flags |= MS_POSIXACL;
343 sbi->ll_flags |= LL_SBI_ACL;
345 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
347 sb->s_flags &= ~MS_POSIXACL;
349 sbi->ll_flags &= ~LL_SBI_ACL;
352 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
353 sbi->ll_flags |= LL_SBI_64BIT_HASH;
355 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
356 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_SHIFT;
358 sbi->ll_md_brw_pages = 1;
360 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
361 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
363 if (obd_connect_has_secctx(data))
364 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
366 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
367 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
368 LCONSOLE_INFO("%s: disabling xattr cache due to "
369 "unknown maximum xattr size.\n", dt);
371 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
372 sbi->ll_xattr_cache_enabled = 1;
376 obd = class_name2obd(dt);
378 CERROR("DT %s: not setup or attached\n", dt);
379 GOTO(out_md_fid, err = -ENODEV);
382 /* pass client page size via ocd_grant_blkbits, the server should report
383 * back its backend blocksize for grant calculation purpose */
384 data->ocd_grant_blkbits = PAGE_SHIFT;
386 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
387 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
388 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
389 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
390 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
391 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
392 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
393 OBD_CONNECT_EINPROGRESS |
394 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
395 OBD_CONNECT_LAYOUTLOCK |
396 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
397 OBD_CONNECT_BULK_MBITS;
399 data->ocd_connect_flags2 = 0;
401 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
402 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
404 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
405 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
406 * disabled by default, because it can still be enabled on the
407 * fly via /proc. As a consequence, we still need to come to an
408 * agreement on the supported algorithms at connect time */
409 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
411 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
412 data->ocd_cksum_types = OBD_CKSUM_ADLER;
414 data->ocd_cksum_types = cksum_types_supported_client();
417 #ifdef HAVE_LRU_RESIZE_SUPPORT
418 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
420 /* always ping even if server suppress_pings */
421 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
422 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
424 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
425 "ocd_grant: %d\n", data->ocd_connect_flags,
426 data->ocd_version, data->ocd_grant);
428 obd->obd_upcall.onu_owner = &sbi->ll_lco;
429 obd->obd_upcall.onu_upcall = cl_ocd_update;
431 data->ocd_brw_size = DT_MAX_BRW_SIZE;
433 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
436 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
437 "recovery, of which this client is not a "
438 "part. Please wait for recovery to "
439 "complete, abort, or time out.\n", dt);
442 CERROR("%s: Cannot connect to %s: rc = %d\n",
443 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
447 sbi->ll_dt_exp->exp_connect_data = *data;
449 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
450 LUSTRE_SEQ_METADATA);
452 CERROR("%s: Can't init data layer FID infrastructure, "
453 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
457 mutex_lock(&sbi->ll_lco.lco_lock);
458 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
459 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
460 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
461 mutex_unlock(&sbi->ll_lco.lco_lock);
463 fid_zero(&sbi->ll_root_fid);
464 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
467 CERROR("cannot mds_connect: rc = %d\n", err);
468 GOTO(out_lock_cn_cb, err);
470 if (!fid_is_sane(&sbi->ll_root_fid)) {
471 CERROR("%s: Invalid root fid "DFID" during mount\n",
472 sbi->ll_md_exp->exp_obd->obd_name,
473 PFID(&sbi->ll_root_fid));
474 GOTO(out_lock_cn_cb, err = -EINVAL);
476 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
478 sb->s_op = &lustre_super_operations;
479 #if THREAD_SIZE >= 8192 /*b=17630*/
480 sb->s_export_op = &lustre_export_operations;
484 * XXX: move this to after cbd setup? */
485 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
486 if (sbi->ll_flags & LL_SBI_ACL)
487 valid |= OBD_MD_FLACL;
489 OBD_ALLOC_PTR(op_data);
491 GOTO(out_lock_cn_cb, err = -ENOMEM);
493 op_data->op_fid1 = sbi->ll_root_fid;
494 op_data->op_mode = 0;
495 op_data->op_valid = valid;
497 err = md_getattr(sbi->ll_md_exp, op_data, &request);
499 OBD_FREE_PTR(op_data);
501 CERROR("%s: md_getattr failed for root: rc = %d\n",
502 sbi->ll_md_exp->exp_obd->obd_name, err);
503 GOTO(out_lock_cn_cb, err);
506 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
507 sbi->ll_md_exp, &lmd);
509 CERROR("failed to understand root inode md: rc = %d\n", err);
510 ptlrpc_req_finished(request);
511 GOTO(out_lock_cn_cb, err);
514 LASSERT(fid_is_sane(&sbi->ll_root_fid));
515 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
516 sbi->ll_flags & LL_SBI_32BIT_API),
518 md_free_lustre_md(sbi->ll_md_exp, &lmd);
519 ptlrpc_req_finished(request);
522 #ifdef CONFIG_FS_POSIX_ACL
524 posix_acl_release(lmd.posix_acl);
525 lmd.posix_acl = NULL;
528 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
530 CERROR("lustre_lite: bad iget4 for root\n");
534 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
535 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
536 KEY_CHECKSUM, sizeof(checksum), &checksum,
539 CERROR("%s: Set checksum failed: rc = %d\n",
540 sbi->ll_dt_exp->exp_obd->obd_name, err);
545 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
546 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
547 sbi->ll_cache, NULL);
549 CERROR("%s: Set cache_set failed: rc = %d\n",
550 sbi->ll_dt_exp->exp_obd->obd_name, err);
554 sb->s_root = d_make_root(root);
555 if (sb->s_root == NULL) {
556 CERROR("%s: can't make root dentry\n",
557 ll_get_fsname(sb, NULL, 0));
558 GOTO(out_root, err = -ENOMEM);
560 #ifdef HAVE_DCACHE_LOCK
561 sb->s_root->d_op = &ll_d_ops;
564 sbi->ll_sdev_orig = sb->s_dev;
566 /* We set sb->s_dev equal on all lustre clients in order to support
567 * NFS export clustering. NFSD requires that the FSID be the same
569 /* s_dev is also used in lt_compare() to compare two fs, but that is
570 * only a node-local comparison. */
571 uuid = obd_get_uuid(sbi->ll_md_exp);
573 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
579 if (proc_lustre_fs_root != NULL) {
580 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
583 CERROR("%s: could not register mount in lprocfs: "
584 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
594 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
596 obd_disconnect(sbi->ll_dt_exp);
597 sbi->ll_dt_exp = NULL;
599 obd_fid_fini(sbi->ll_md_exp->exp_obd);
601 obd_disconnect(sbi->ll_md_exp);
602 sbi->ll_md_exp = NULL;
611 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
615 size = sizeof(*lmmsize);
616 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
617 KEY_MAX_EASIZE, &size, lmmsize);
619 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
620 sbi->ll_dt_exp->exp_obd->obd_name, rc);
625 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
626 KEY_MAX_EASIZE, &size, lmmsize);
628 CERROR("Get max mdsize error rc %d\n", rc);
634 * Get the value of the default_easize parameter.
636 * \see client_obd::cl_default_mds_easize
638 * \param[in] sbi superblock info for this filesystem
639 * \param[out] lmmsize pointer to storage location for value
641 * \retval 0 on success
642 * \retval negative negated errno on failure
644 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
649 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
650 KEY_DEFAULT_EASIZE, &size, lmmsize);
652 CERROR("Get default mdsize error rc %d\n", rc);
658 * Set the default_easize parameter to the given value.
660 * \see client_obd::cl_default_mds_easize
662 * \param[in] sbi superblock info for this filesystem
663 * \param[in] lmmsize the size to set
665 * \retval 0 on success
666 * \retval negative negated errno on failure
668 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
672 if (lmmsize < sizeof(struct lov_mds_md) ||
673 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
676 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
677 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
678 sizeof(int), &lmmsize, NULL);
683 static void ll_dump_inode(struct inode *inode)
685 struct ll_d_hlist_node *tmp;
686 int dentry_count = 0;
688 LASSERT(inode != NULL);
690 ll_d_hlist_for_each(tmp, &inode->i_dentry)
693 CERROR("%s: inode %p dump: dev=%s fid="DFID
694 " mode=%o count=%u, %d dentries\n",
695 ll_get_fsname(inode->i_sb, NULL, 0), inode,
696 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
697 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
700 void lustre_dump_dentry(struct dentry *dentry, int recur)
702 struct list_head *tmp;
705 LASSERT(dentry != NULL);
707 list_for_each(tmp, &dentry->d_subdirs)
710 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
711 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
712 dentry->d_name.len, dentry->d_name.name,
713 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
714 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
715 dentry->d_flags, dentry->d_fsdata, subdirs);
716 if (dentry->d_inode != NULL)
717 ll_dump_inode(dentry->d_inode);
722 list_for_each(tmp, &dentry->d_subdirs) {
723 struct dentry *d = list_entry(tmp, struct dentry, d_child);
724 lustre_dump_dentry(d, recur - 1);
728 static void client_common_put_super(struct super_block *sb)
730 struct ll_sb_info *sbi = ll_s2sbi(sb);
735 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
736 obd_disconnect(sbi->ll_dt_exp);
737 sbi->ll_dt_exp = NULL;
739 lprocfs_unregister_mountpoint(sbi);
741 obd_fid_fini(sbi->ll_md_exp->exp_obd);
742 obd_disconnect(sbi->ll_md_exp);
743 sbi->ll_md_exp = NULL;
748 void ll_kill_super(struct super_block *sb)
750 struct ll_sb_info *sbi;
754 if (!(sb->s_flags & MS_ACTIVE))
758 /* we need restore s_dev from changed for clustred NFS before put_super
759 * because new kernels have cached s_dev and change sb->s_dev in
760 * put_super not affected real removing devices */
762 sb->s_dev = sbi->ll_sdev_orig;
763 sbi->ll_umounting = 1;
765 /* wait running statahead threads to quit */
766 while (atomic_read(&sbi->ll_sa_running) > 0) {
767 set_current_state(TASK_UNINTERRUPTIBLE);
768 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
775 static inline int ll_set_opt(const char *opt, char *data, int fl)
777 if (strncmp(opt, data, strlen(opt)) != 0)
783 /* non-client-specific mount options are parsed in lmd_parse */
784 static int ll_options(char *options, int *flags)
787 char *s1 = options, *s2;
793 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
796 CDEBUG(D_SUPER, "next opt=%s\n", s1);
797 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
802 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
807 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
812 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
817 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
822 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
827 tmp = ll_set_opt("context", s1, 1);
830 tmp = ll_set_opt("fscontext", s1, 1);
833 tmp = ll_set_opt("defcontext", s1, 1);
836 tmp = ll_set_opt("rootcontext", s1, 1);
839 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
844 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
850 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
855 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
860 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
865 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
870 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
875 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
880 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
885 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
890 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
895 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
900 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
906 s2 = strchr(s1, ',');
914 void ll_lli_init(struct ll_inode_info *lli)
916 lli->lli_inode_magic = LLI_INODE_MAGIC;
918 spin_lock_init(&lli->lli_lock);
919 lli->lli_posix_acl = NULL;
920 /* Do not set lli_fid, it has been initialized already. */
921 fid_zero(&lli->lli_pfid);
922 lli->lli_mds_read_och = NULL;
923 lli->lli_mds_write_och = NULL;
924 lli->lli_mds_exec_och = NULL;
925 lli->lli_open_fd_read_count = 0;
926 lli->lli_open_fd_write_count = 0;
927 lli->lli_open_fd_exec_count = 0;
928 mutex_init(&lli->lli_och_mutex);
929 spin_lock_init(&lli->lli_agl_lock);
930 spin_lock_init(&lli->lli_layout_lock);
931 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
932 lli->lli_clob = NULL;
934 init_rwsem(&lli->lli_xattrs_list_rwsem);
935 mutex_init(&lli->lli_xattrs_enq_lock);
937 LASSERT(lli->lli_vfs_inode.i_mode != 0);
938 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
939 mutex_init(&lli->lli_readdir_mutex);
940 lli->lli_opendir_key = NULL;
942 spin_lock_init(&lli->lli_sa_lock);
943 lli->lli_opendir_pid = 0;
944 lli->lli_sa_enabled = 0;
945 lli->lli_def_stripe_offset = -1;
947 mutex_init(&lli->lli_size_mutex);
948 lli->lli_symlink_name = NULL;
949 init_rwsem(&lli->lli_trunc_sem);
950 range_lock_tree_init(&lli->lli_write_tree);
951 init_rwsem(&lli->lli_glimpse_sem);
952 lli->lli_glimpse_time = 0;
953 INIT_LIST_HEAD(&lli->lli_agl_list);
954 lli->lli_agl_index = 0;
955 lli->lli_async_rc = 0;
957 mutex_init(&lli->lli_layout_mutex);
960 static inline int ll_bdi_register(struct backing_dev_info *bdi)
962 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
964 bdi->name = "lustre";
965 return bdi_register(bdi, NULL, "lustre-%d",
966 atomic_inc_return(&ll_bdi_num));
969 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
971 struct lustre_profile *lprof = NULL;
972 struct lustre_sb_info *lsi = s2lsi(sb);
973 struct ll_sb_info *sbi;
974 char *dt = NULL, *md = NULL;
975 char *profilenm = get_profile_name(sb);
976 struct config_llog_instance *cfg;
977 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
978 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
982 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
988 try_module_get(THIS_MODULE);
990 /* client additional sb info */
991 lsi->lsi_llsbi = sbi = ll_init_sbi();
993 module_put(THIS_MODULE);
998 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1000 GOTO(out_free, err);
1002 err = bdi_init(&lsi->lsi_bdi);
1004 GOTO(out_free, err);
1005 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1006 #ifdef HAVE_BDI_CAP_MAP_COPY
1007 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1009 lsi->lsi_bdi.capabilities = 0;
1011 err = ll_bdi_register(&lsi->lsi_bdi);
1013 GOTO(out_free, err);
1015 sb->s_bdi = &lsi->lsi_bdi;
1016 #ifndef HAVE_DCACHE_LOCK
1017 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1018 sb->s_d_op = &ll_d_ops;
1021 /* Generate a string unique to this super, in case some joker tries
1022 to mount the same fs at two mount points.
1023 Use the address of the super itself.*/
1024 cfg->cfg_instance = sb;
1025 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1026 cfg->cfg_callback = class_config_llog_handler;
1027 /* set up client obds */
1028 err = lustre_process_log(sb, profilenm, cfg);
1030 GOTO(out_free, err);
1032 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1033 lprof = class_get_profile(profilenm);
1034 if (lprof == NULL) {
1035 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1036 " read from the MGS. Does that filesystem "
1037 "exist?\n", profilenm);
1038 GOTO(out_free, err = -EINVAL);
1040 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1041 lprof->lp_md, lprof->lp_dt);
1043 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1045 GOTO(out_free, err = -ENOMEM);
1046 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1048 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1050 GOTO(out_free, err = -ENOMEM);
1051 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1053 /* connections, registrations, sb setup */
1054 err = client_common_fill_super(sb, md, dt, mnt);
1056 GOTO(out_free, err);
1058 sbi->ll_client_common_fill_super_succeeded = 1;
1062 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1064 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1066 class_put_profile(lprof);
1069 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1070 LCONSOLE_WARN("Mounted %s\n", profilenm);
1074 } /* ll_fill_super */
1076 void ll_put_super(struct super_block *sb)
1078 struct config_llog_instance cfg, params_cfg;
1079 struct obd_device *obd;
1080 struct lustre_sb_info *lsi = s2lsi(sb);
1081 struct ll_sb_info *sbi = ll_s2sbi(sb);
1082 char *profilenm = get_profile_name(sb);
1084 int next, force = 1, rc = 0;
1087 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1089 cfg.cfg_instance = sb;
1090 lustre_end_log(sb, profilenm, &cfg);
1092 params_cfg.cfg_instance = sb;
1093 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1095 if (sbi->ll_md_exp) {
1096 obd = class_exp2obd(sbi->ll_md_exp);
1098 force = obd->obd_force;
1101 /* Wait for unstable pages to be committed to stable storage */
1103 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1104 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1105 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1109 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1110 if (force == 0 && rc != -EINTR)
1111 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1114 /* We need to set force before the lov_disconnect in
1115 lustre_common_put_super, since l_d cleans up osc's as well. */
1118 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1120 obd->obd_force = force;
1124 if (sbi->ll_client_common_fill_super_succeeded) {
1125 /* Only if client_common_fill_super succeeded */
1126 client_common_put_super(sb);
1130 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1131 class_manual_cleanup(obd);
1134 if (sbi->ll_flags & LL_SBI_VERBOSE)
1135 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1138 class_del_profile(profilenm);
1140 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1141 bdi_destroy(&lsi->lsi_bdi);
1142 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1146 lsi->lsi_llsbi = NULL;
1148 lustre_common_put_super(sb);
1150 cl_env_cache_purge(~0);
1152 module_put(THIS_MODULE);
1155 } /* client_put_super */
1157 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1159 struct inode *inode = NULL;
1161 /* NOTE: we depend on atomic igrab() -bzzz */
1162 lock_res_and_lock(lock);
1163 if (lock->l_resource->lr_lvb_inode) {
1164 struct ll_inode_info * lli;
1165 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1166 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1167 inode = igrab(lock->l_resource->lr_lvb_inode);
1169 inode = lock->l_resource->lr_lvb_inode;
1170 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1171 D_WARNING, lock, "lr_lvb_inode %p is "
1172 "bogus: magic %08x",
1173 lock->l_resource->lr_lvb_inode,
1174 lli->lli_inode_magic);
1178 unlock_res_and_lock(lock);
1182 void ll_dir_clear_lsm_md(struct inode *inode)
1184 struct ll_inode_info *lli = ll_i2info(inode);
1186 LASSERT(S_ISDIR(inode->i_mode));
1188 if (lli->lli_lsm_md != NULL) {
1189 lmv_free_memmd(lli->lli_lsm_md);
1190 lli->lli_lsm_md = NULL;
1194 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1195 const struct lu_fid *fid,
1196 struct lustre_md *md)
1198 struct ll_sb_info *sbi = ll_s2sbi(sb);
1199 struct mdt_body *body = md->body;
1200 struct inode *inode;
1204 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1205 inode = iget_locked(sb, ino);
1206 if (inode == NULL) {
1207 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1208 ll_get_fsname(sb, NULL, 0), PFID(fid));
1209 RETURN(ERR_PTR(-ENOENT));
1212 if (inode->i_state & I_NEW) {
1213 struct ll_inode_info *lli = ll_i2info(inode);
1214 struct lmv_stripe_md *lsm = md->lmv;
1216 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1217 (body->mbo_mode & S_IFMT);
1218 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1221 LTIME_S(inode->i_mtime) = 0;
1222 LTIME_S(inode->i_atime) = 0;
1223 LTIME_S(inode->i_ctime) = 0;
1226 #ifdef HAVE_BACKING_DEV_INFO
1227 /* initializing backing dev info. */
1228 inode->i_mapping->backing_dev_info =
1229 &s2lsi(inode->i_sb)->lsi_bdi;
1231 inode->i_op = &ll_dir_inode_operations;
1232 inode->i_fop = &ll_dir_operations;
1233 lli->lli_fid = *fid;
1236 LASSERT(lsm != NULL);
1237 /* master object FID */
1238 lli->lli_pfid = body->mbo_fid1;
1239 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1240 lli, PFID(fid), PFID(&lli->lli_pfid));
1241 unlock_new_inode(inode);
1247 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1250 struct lmv_stripe_md *lsm = md->lmv;
1253 LASSERT(lsm != NULL);
1254 /* XXX sigh, this lsm_root initialization should be in
1255 * LMV layer, but it needs ll_iget right now, so we
1256 * put this here right now. */
1257 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1258 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1259 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1260 /* Unfortunately ll_iget will call ll_update_inode,
1261 * where the initialization of slave inode is slightly
1262 * different, so it reset lsm_md to NULL to avoid
1263 * initializing lsm for slave inode. */
1264 /* For migrating inode, master stripe and master object will
1265 * be same, so we only need assign this inode */
1266 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1267 lsm->lsm_md_oinfo[i].lmo_root = inode;
1269 lsm->lsm_md_oinfo[i].lmo_root =
1270 ll_iget_anon_dir(inode->i_sb, fid, md);
1272 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1273 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1275 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1283 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1284 const struct lmv_stripe_md *lsm_md2)
1286 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1287 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1288 lsm_md1->lsm_md_master_mdt_index ==
1289 lsm_md2->lsm_md_master_mdt_index &&
1290 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1291 lsm_md1->lsm_md_layout_version ==
1292 lsm_md2->lsm_md_layout_version &&
1293 strcmp(lsm_md1->lsm_md_pool_name,
1294 lsm_md2->lsm_md_pool_name) == 0;
1297 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1299 struct ll_inode_info *lli = ll_i2info(inode);
1300 struct lmv_stripe_md *lsm = md->lmv;
1304 LASSERT(S_ISDIR(inode->i_mode));
1305 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1306 PFID(ll_inode2fid(inode)));
1308 /* no striped information from request. */
1310 if (lli->lli_lsm_md == NULL) {
1312 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1313 LMV_HASH_FLAG_MIGRATION) {
1314 /* migration is done, the temporay MIGRATE layout has
1316 CDEBUG(D_INODE, DFID" finish migration.\n",
1317 PFID(ll_inode2fid(inode)));
1318 lmv_free_memmd(lli->lli_lsm_md);
1319 lli->lli_lsm_md = NULL;
1322 /* The lustre_md from req does not include stripeEA,
1323 * see ll_md_setattr */
1328 /* set the directory layout */
1329 if (lli->lli_lsm_md == NULL) {
1330 struct cl_attr *attr;
1332 rc = ll_init_lsm_md(inode, md);
1336 /* set md->lmv to NULL, so the following free lustre_md
1337 * will not free this lsm */
1339 lli->lli_lsm_md = lsm;
1341 OBD_ALLOC_PTR(attr);
1345 /* validate the lsm */
1346 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1347 ll_md_blocking_ast);
1353 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1354 md->body->mbo_nlink = attr->cat_nlink;
1355 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1356 md->body->mbo_size = attr->cat_size;
1357 if (md->body->mbo_valid & OBD_MD_FLATIME)
1358 md->body->mbo_atime = attr->cat_atime;
1359 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1360 md->body->mbo_ctime = attr->cat_ctime;
1361 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1362 md->body->mbo_mtime = attr->cat_mtime;
1366 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1367 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1371 /* Compare the old and new stripe information */
1372 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1373 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1376 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1377 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1378 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1379 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1380 inode, lsm, old_lsm,
1381 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1382 lsm->lsm_md_stripe_count,
1383 old_lsm->lsm_md_stripe_count,
1384 lsm->lsm_md_master_mdt_index,
1385 old_lsm->lsm_md_master_mdt_index,
1386 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1387 lsm->lsm_md_layout_version,
1388 old_lsm->lsm_md_layout_version,
1389 lsm->lsm_md_pool_name,
1390 old_lsm->lsm_md_pool_name);
1392 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1393 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1394 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1395 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1398 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1399 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1400 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1401 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1410 void ll_clear_inode(struct inode *inode)
1412 struct ll_inode_info *lli = ll_i2info(inode);
1413 struct ll_sb_info *sbi = ll_i2sbi(inode);
1416 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1417 PFID(ll_inode2fid(inode)), inode);
1419 if (S_ISDIR(inode->i_mode)) {
1420 /* these should have been cleared in ll_file_release */
1421 LASSERT(lli->lli_opendir_key == NULL);
1422 LASSERT(lli->lli_sai == NULL);
1423 LASSERT(lli->lli_opendir_pid == 0);
1426 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1428 LASSERT(!lli->lli_open_fd_write_count);
1429 LASSERT(!lli->lli_open_fd_read_count);
1430 LASSERT(!lli->lli_open_fd_exec_count);
1432 if (lli->lli_mds_write_och)
1433 ll_md_real_close(inode, FMODE_WRITE);
1434 if (lli->lli_mds_exec_och)
1435 ll_md_real_close(inode, FMODE_EXEC);
1436 if (lli->lli_mds_read_och)
1437 ll_md_real_close(inode, FMODE_READ);
1439 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1440 OBD_FREE(lli->lli_symlink_name,
1441 strlen(lli->lli_symlink_name) + 1);
1442 lli->lli_symlink_name = NULL;
1445 ll_xattr_cache_destroy(inode);
1447 #ifdef CONFIG_FS_POSIX_ACL
1448 if (lli->lli_posix_acl) {
1449 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1450 posix_acl_release(lli->lli_posix_acl);
1451 lli->lli_posix_acl = NULL;
1454 lli->lli_inode_magic = LLI_INODE_DEAD;
1456 if (S_ISDIR(inode->i_mode))
1457 ll_dir_clear_lsm_md(inode);
1458 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1459 LASSERT(list_empty(&lli->lli_agl_list));
1462 * XXX This has to be done before lsm is freed below, because
1463 * cl_object still uses inode lsm.
1465 cl_inode_fini(inode);
1470 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1472 struct lustre_md md;
1473 struct inode *inode = dentry->d_inode;
1474 struct ll_sb_info *sbi = ll_i2sbi(inode);
1475 struct ptlrpc_request *request = NULL;
1479 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1480 LUSTRE_OPC_ANY, NULL);
1481 if (IS_ERR(op_data))
1482 RETURN(PTR_ERR(op_data));
1484 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1486 ptlrpc_req_finished(request);
1487 if (rc == -ENOENT) {
1489 /* Unlinked special device node? Or just a race?
1490 * Pretend we done everything. */
1491 if (!S_ISREG(inode->i_mode) &&
1492 !S_ISDIR(inode->i_mode)) {
1493 ia_valid = op_data->op_attr.ia_valid;
1494 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1495 rc = simple_setattr(dentry, &op_data->op_attr);
1496 op_data->op_attr.ia_valid = ia_valid;
1498 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1499 CERROR("md_setattr fails: rc = %d\n", rc);
1504 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1505 sbi->ll_md_exp, &md);
1507 ptlrpc_req_finished(request);
1511 ia_valid = op_data->op_attr.ia_valid;
1512 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1513 * cache is not cleared yet. */
1514 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1515 if (S_ISREG(inode->i_mode))
1517 rc = simple_setattr(dentry, &op_data->op_attr);
1518 if (S_ISREG(inode->i_mode))
1519 inode_unlock(inode);
1520 op_data->op_attr.ia_valid = ia_valid;
1522 rc = ll_update_inode(inode, &md);
1523 ptlrpc_req_finished(request);
1528 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1529 * object(s) determine the file size and mtime. Otherwise, the MDS will
1530 * keep these values until such a time that objects are allocated for it.
1531 * We do the MDS operations first, as it is checking permissions for us.
1532 * We don't to the MDS RPC if there is nothing that we want to store there,
1533 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1534 * going to do an RPC anyways.
1536 * If we are doing a truncate, we will send the mtime and ctime updates
1537 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1538 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1541 * In case of HSMimport, we only set attr on MDS.
1543 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1545 struct inode *inode = dentry->d_inode;
1546 struct ll_inode_info *lli = ll_i2info(inode);
1547 struct md_op_data *op_data = NULL;
1551 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1552 "valid %x, hsm_import %d\n",
1553 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1554 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1557 if (attr->ia_valid & ATTR_SIZE) {
1558 /* Check new size against VFS/VM file size limit and rlimit */
1559 rc = inode_newsize_ok(inode, attr->ia_size);
1563 /* The maximum Lustre file size is variable, based on the
1564 * OST maximum object size and number of stripes. This
1565 * needs another check in addition to the VFS check above. */
1566 if (attr->ia_size > ll_file_maxbytes(inode)) {
1567 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1568 PFID(&lli->lli_fid), attr->ia_size,
1569 ll_file_maxbytes(inode));
1573 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1576 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1577 if (attr->ia_valid & TIMES_SET_FLAGS) {
1578 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1579 !cfs_capable(CFS_CAP_FOWNER))
1583 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1584 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1585 (attr->ia_valid & ATTR_CTIME)) {
1586 attr->ia_ctime = CURRENT_TIME;
1587 attr->ia_valid |= ATTR_CTIME_SET;
1589 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1590 (attr->ia_valid & ATTR_ATIME)) {
1591 attr->ia_atime = CURRENT_TIME;
1592 attr->ia_valid |= ATTR_ATIME_SET;
1594 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1595 (attr->ia_valid & ATTR_MTIME)) {
1596 attr->ia_mtime = CURRENT_TIME;
1597 attr->ia_valid |= ATTR_MTIME_SET;
1600 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1601 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1602 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1603 cfs_time_current_sec());
1605 if (S_ISREG(inode->i_mode)) {
1606 if (attr->ia_valid & ATTR_SIZE)
1607 inode_dio_write_done(inode);
1608 inode_unlock(inode);
1611 /* We always do an MDS RPC, even if we're only changing the size;
1612 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1614 OBD_ALLOC_PTR(op_data);
1615 if (op_data == NULL)
1616 GOTO(out, rc = -ENOMEM);
1618 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1619 /* If we are changing file size, file content is
1620 * modified, flag it. */
1621 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1622 op_data->op_bias |= MDS_DATA_MODIFIED;
1623 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1626 op_data->op_attr = *attr;
1628 rc = ll_md_setattr(dentry, op_data);
1632 if (!S_ISREG(inode->i_mode) || hsm_import)
1635 if (attr->ia_valid & (ATTR_SIZE |
1636 ATTR_ATIME | ATTR_ATIME_SET |
1637 ATTR_MTIME | ATTR_MTIME_SET |
1638 ATTR_CTIME | ATTR_CTIME_SET)) {
1639 /* For truncate and utimes sending attributes to OSTs, setting
1640 * mtime/atime to the past will be performed under PW [0:EOF]
1641 * extent lock (new_size:EOF for truncate). It may seem
1642 * excessive to send mtime/atime updates to OSTs when not
1643 * setting times to past, but it is necessary due to possible
1644 * time de-synchronization between MDT inode and OST objects */
1645 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1648 /* If the file was restored, it needs to set dirty flag.
1650 * We've already sent MDS_DATA_MODIFIED flag in
1651 * ll_md_setattr() for truncate. However, the MDT refuses to
1652 * set the HS_DIRTY flag on released files, so we have to set
1653 * it again if the file has been restored. Please check how
1654 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1656 * Please notice that if the file is not released, the previous
1657 * MDS_DATA_MODIFIED has taken effect and usually
1658 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1659 * This way we can save an RPC for common open + trunc
1661 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1662 struct hsm_state_set hss = {
1663 .hss_valid = HSS_SETMASK,
1664 .hss_setmask = HS_DIRTY,
1668 rc2 = ll_hsm_state_set(inode, &hss);
1669 /* truncate and write can happen at the same time, so that
1670 * the file can be set modified even though the file is not
1671 * restored from released state, and ll_hsm_state_set() is
1672 * not applicable for the file, and rc2 < 0 is normal in this
1675 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1676 PFID(ll_inode2fid(inode)), rc2);
1681 if (op_data != NULL)
1682 ll_finish_md_op_data(op_data);
1684 if (S_ISREG(inode->i_mode)) {
1686 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1687 inode_dio_wait(inode);
1690 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1691 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1696 int ll_setattr(struct dentry *de, struct iattr *attr)
1698 int mode = de->d_inode->i_mode;
1700 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1701 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1702 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1704 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1705 (ATTR_SIZE|ATTR_MODE)) &&
1706 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1707 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1708 !(attr->ia_mode & S_ISGID))))
1709 attr->ia_valid |= ATTR_FORCE;
1711 if ((attr->ia_valid & ATTR_MODE) &&
1713 !(attr->ia_mode & S_ISUID) &&
1714 !(attr->ia_valid & ATTR_KILL_SUID))
1715 attr->ia_valid |= ATTR_KILL_SUID;
1717 if ((attr->ia_valid & ATTR_MODE) &&
1718 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1719 !(attr->ia_mode & S_ISGID) &&
1720 !(attr->ia_valid & ATTR_KILL_SGID))
1721 attr->ia_valid |= ATTR_KILL_SGID;
1723 return ll_setattr_raw(de, attr, false);
1726 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1727 __u64 max_age, __u32 flags)
1729 struct ll_sb_info *sbi = ll_s2sbi(sb);
1730 struct obd_statfs obd_osfs;
1734 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1736 CERROR("md_statfs fails: rc = %d\n", rc);
1740 osfs->os_type = sb->s_magic;
1742 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1743 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1745 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1746 flags |= OBD_STATFS_NODELAY;
1748 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1750 CERROR("obd_statfs fails: rc = %d\n", rc);
1754 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1755 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1758 osfs->os_bsize = obd_osfs.os_bsize;
1759 osfs->os_blocks = obd_osfs.os_blocks;
1760 osfs->os_bfree = obd_osfs.os_bfree;
1761 osfs->os_bavail = obd_osfs.os_bavail;
1763 /* If we don't have as many objects free on the OST as inodes
1764 * on the MDS, we reduce the total number of inodes to
1765 * compensate, so that the "inodes in use" number is correct.
1767 if (obd_osfs.os_ffree < osfs->os_ffree) {
1768 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1770 osfs->os_ffree = obd_osfs.os_ffree;
1775 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1777 struct super_block *sb = de->d_sb;
1778 struct obd_statfs osfs;
1779 __u64 fsid = huge_encode_dev(sb->s_dev);
1782 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1783 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1785 /* Some amount of caching on the client is allowed */
1786 rc = ll_statfs_internal(sb, &osfs,
1787 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1792 statfs_unpack(sfs, &osfs);
1794 /* We need to downshift for all 32-bit kernels, because we can't
1795 * tell if the kernel is being called via sys_statfs64() or not.
1796 * Stop before overflowing f_bsize - in which case it is better
1797 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1798 if (sizeof(long) < 8) {
1799 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1802 osfs.os_blocks >>= 1;
1803 osfs.os_bfree >>= 1;
1804 osfs.os_bavail >>= 1;
1808 sfs->f_blocks = osfs.os_blocks;
1809 sfs->f_bfree = osfs.os_bfree;
1810 sfs->f_bavail = osfs.os_bavail;
1811 sfs->f_fsid.val[0] = (__u32)fsid;
1812 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1816 void ll_inode_size_lock(struct inode *inode)
1818 struct ll_inode_info *lli;
1820 LASSERT(!S_ISDIR(inode->i_mode));
1822 lli = ll_i2info(inode);
1823 mutex_lock(&lli->lli_size_mutex);
1826 void ll_inode_size_unlock(struct inode *inode)
1828 struct ll_inode_info *lli;
1830 lli = ll_i2info(inode);
1831 mutex_unlock(&lli->lli_size_mutex);
1834 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1836 struct ll_inode_info *lli = ll_i2info(inode);
1837 struct mdt_body *body = md->body;
1838 struct ll_sb_info *sbi = ll_i2sbi(inode);
1840 if (body->mbo_valid & OBD_MD_FLEASIZE)
1841 cl_file_inode_init(inode, md);
1843 if (S_ISDIR(inode->i_mode)) {
1846 rc = ll_update_lsm_md(inode, md);
1851 #ifdef CONFIG_FS_POSIX_ACL
1852 if (body->mbo_valid & OBD_MD_FLACL) {
1853 spin_lock(&lli->lli_lock);
1854 if (lli->lli_posix_acl)
1855 posix_acl_release(lli->lli_posix_acl);
1856 lli->lli_posix_acl = md->posix_acl;
1857 spin_unlock(&lli->lli_lock);
1860 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1861 sbi->ll_flags & LL_SBI_32BIT_API);
1862 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1864 if (body->mbo_valid & OBD_MD_FLATIME) {
1865 if (body->mbo_atime > LTIME_S(inode->i_atime))
1866 LTIME_S(inode->i_atime) = body->mbo_atime;
1867 lli->lli_atime = body->mbo_atime;
1870 if (body->mbo_valid & OBD_MD_FLMTIME) {
1871 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1872 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1873 "to %llu\n", inode->i_ino,
1874 LTIME_S(inode->i_mtime), body->mbo_mtime);
1875 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1877 lli->lli_mtime = body->mbo_mtime;
1880 if (body->mbo_valid & OBD_MD_FLCTIME) {
1881 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1882 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1883 lli->lli_ctime = body->mbo_ctime;
1886 if (body->mbo_valid & OBD_MD_FLMODE)
1887 inode->i_mode = (inode->i_mode & S_IFMT) |
1888 (body->mbo_mode & ~S_IFMT);
1890 if (body->mbo_valid & OBD_MD_FLTYPE)
1891 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1892 (body->mbo_mode & S_IFMT);
1894 LASSERT(inode->i_mode != 0);
1895 if (S_ISREG(inode->i_mode))
1896 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1897 LL_MAX_BLKSIZE_BITS);
1899 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1901 if (body->mbo_valid & OBD_MD_FLUID)
1902 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1903 if (body->mbo_valid & OBD_MD_FLGID)
1904 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1905 if (body->mbo_valid & OBD_MD_FLFLAGS)
1906 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1907 if (body->mbo_valid & OBD_MD_FLNLINK)
1908 set_nlink(inode, body->mbo_nlink);
1909 if (body->mbo_valid & OBD_MD_FLRDEV)
1910 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1912 if (body->mbo_valid & OBD_MD_FLID) {
1913 /* FID shouldn't be changed! */
1914 if (fid_is_sane(&lli->lli_fid)) {
1915 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1916 "Trying to change FID "DFID
1917 " to the "DFID", inode "DFID"(%p)\n",
1918 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1919 PFID(ll_inode2fid(inode)), inode);
1921 lli->lli_fid = body->mbo_fid1;
1925 LASSERT(fid_seq(&lli->lli_fid) != 0);
1927 if (body->mbo_valid & OBD_MD_FLSIZE) {
1928 i_size_write(inode, body->mbo_size);
1930 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1931 PFID(ll_inode2fid(inode)),
1932 (unsigned long long)body->mbo_size);
1934 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1935 inode->i_blocks = body->mbo_blocks;
1938 if (body->mbo_valid & OBD_MD_TSTATE) {
1939 /* Set LLIF_FILE_RESTORING if restore ongoing and
1940 * clear it when done to ensure to start again
1941 * glimpsing updated attrs
1943 if (body->mbo_t_state & MS_RESTORE)
1944 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1946 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1952 int ll_read_inode2(struct inode *inode, void *opaque)
1954 struct lustre_md *md = opaque;
1955 struct ll_inode_info *lli = ll_i2info(inode);
1959 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1960 PFID(&lli->lli_fid), inode);
1962 /* Core attributes from the MDS first. This is a new inode, and
1963 * the VFS doesn't zero times in the core inode so we have to do
1964 * it ourselves. They will be overwritten by either MDS or OST
1965 * attributes - we just need to make sure they aren't newer. */
1966 LTIME_S(inode->i_mtime) = 0;
1967 LTIME_S(inode->i_atime) = 0;
1968 LTIME_S(inode->i_ctime) = 0;
1970 rc = ll_update_inode(inode, md);
1974 /* OIDEBUG(inode); */
1976 #ifdef HAVE_BACKING_DEV_INFO
1977 /* initializing backing dev info. */
1978 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1980 if (S_ISREG(inode->i_mode)) {
1981 struct ll_sb_info *sbi = ll_i2sbi(inode);
1982 inode->i_op = &ll_file_inode_operations;
1983 inode->i_fop = sbi->ll_fop;
1984 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1986 } else if (S_ISDIR(inode->i_mode)) {
1987 inode->i_op = &ll_dir_inode_operations;
1988 inode->i_fop = &ll_dir_operations;
1990 } else if (S_ISLNK(inode->i_mode)) {
1991 inode->i_op = &ll_fast_symlink_inode_operations;
1994 inode->i_op = &ll_special_inode_operations;
1996 init_special_inode(inode, inode->i_mode,
2005 void ll_delete_inode(struct inode *inode)
2007 struct ll_inode_info *lli = ll_i2info(inode);
2010 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2011 /* It is last chance to write out dirty pages,
2012 * otherwise we may lose data while umount */
2013 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2015 truncate_inode_pages_final(&inode->i_data);
2017 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2018 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2019 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2021 #ifdef HAVE_SBOPS_EVICT_INODE
2022 ll_clear_inode(inode);
2029 int ll_iocontrol(struct inode *inode, struct file *file,
2030 unsigned int cmd, unsigned long arg)
2032 struct ll_sb_info *sbi = ll_i2sbi(inode);
2033 struct ptlrpc_request *req = NULL;
2038 case FSFILT_IOC_GETFLAGS: {
2039 struct mdt_body *body;
2040 struct md_op_data *op_data;
2042 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2043 0, 0, LUSTRE_OPC_ANY,
2045 if (IS_ERR(op_data))
2046 RETURN(PTR_ERR(op_data));
2048 op_data->op_valid = OBD_MD_FLFLAGS;
2049 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2050 ll_finish_md_op_data(op_data);
2052 CERROR("%s: failure inode "DFID": rc = %d\n",
2053 sbi->ll_md_exp->exp_obd->obd_name,
2054 PFID(ll_inode2fid(inode)), rc);
2058 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2060 flags = body->mbo_flags;
2062 ptlrpc_req_finished(req);
2064 RETURN(put_user(flags, (int __user *)arg));
2066 case FSFILT_IOC_SETFLAGS: {
2068 struct md_op_data *op_data;
2069 struct cl_object *obj;
2071 if (get_user(flags, (int __user *)arg))
2074 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2075 LUSTRE_OPC_ANY, NULL);
2076 if (IS_ERR(op_data))
2077 RETURN(PTR_ERR(op_data));
2079 op_data->op_attr_flags = flags;
2080 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2081 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2082 ll_finish_md_op_data(op_data);
2083 ptlrpc_req_finished(req);
2087 inode->i_flags = ll_ext_to_inode_flags(flags);
2089 obj = ll_i2info(inode)->lli_clob;
2093 OBD_ALLOC_PTR(attr);
2097 attr->ia_valid = ATTR_ATTR_FLAG;
2098 rc = cl_setattr_ost(obj, attr, flags);
2110 int ll_flush_ctx(struct inode *inode)
2112 struct ll_sb_info *sbi = ll_i2sbi(inode);
2114 CDEBUG(D_SEC, "flush context for user %d\n",
2115 from_kuid(&init_user_ns, current_uid()));
2117 obd_set_info_async(NULL, sbi->ll_md_exp,
2118 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2120 obd_set_info_async(NULL, sbi->ll_dt_exp,
2121 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2126 /* umount -f client means force down, don't save state */
2127 void ll_umount_begin(struct super_block *sb)
2129 struct ll_sb_info *sbi = ll_s2sbi(sb);
2130 struct obd_device *obd;
2131 struct obd_ioctl_data *ioc_data;
2132 struct l_wait_info lwi;
2133 wait_queue_head_t waitq;
2136 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2137 sb->s_count, atomic_read(&sb->s_active));
2139 obd = class_exp2obd(sbi->ll_md_exp);
2141 CERROR("Invalid MDC connection handle %#llx\n",
2142 sbi->ll_md_exp->exp_handle.h_cookie);
2148 obd = class_exp2obd(sbi->ll_dt_exp);
2150 CERROR("Invalid LOV connection handle %#llx\n",
2151 sbi->ll_dt_exp->exp_handle.h_cookie);
2157 OBD_ALLOC_PTR(ioc_data);
2159 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2160 sizeof *ioc_data, ioc_data, NULL);
2162 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2163 sizeof *ioc_data, ioc_data, NULL);
2165 OBD_FREE_PTR(ioc_data);
2168 /* Really, we'd like to wait until there are no requests outstanding,
2169 * and then continue. For now, we just periodically checking for vfs
2170 * to decrement mnt_cnt and hope to finish it within 10sec.
2172 init_waitqueue_head(&waitq);
2173 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2174 cfs_time_seconds(1), NULL, NULL);
2175 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2180 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2182 struct ll_sb_info *sbi = ll_s2sbi(sb);
2183 char *profilenm = get_profile_name(sb);
2187 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2188 read_only = *flags & MS_RDONLY;
2189 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2190 sizeof(KEY_READ_ONLY),
2191 KEY_READ_ONLY, sizeof(read_only),
2194 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2195 profilenm, read_only ?
2196 "read-only" : "read-write", err);
2201 sb->s_flags |= MS_RDONLY;
2203 sb->s_flags &= ~MS_RDONLY;
2205 if (sbi->ll_flags & LL_SBI_VERBOSE)
2206 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2207 read_only ? "read-only" : "read-write");
2213 * Cleanup the open handle that is cached on MDT-side.
2215 * For open case, the client side open handling thread may hit error
2216 * after the MDT grant the open. Under such case, the client should
2217 * send close RPC to the MDT as cleanup; otherwise, the open handle
2218 * on the MDT will be leaked there until the client umount or evicted.
2220 * In further, if someone unlinked the file, because the open handle
2221 * holds the reference on such file/object, then it will block the
2222 * subsequent threads that want to locate such object via FID.
2224 * \param[in] sb super block for this file-system
2225 * \param[in] open_req pointer to the original open request
2227 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2229 struct mdt_body *body;
2230 struct md_op_data *op_data;
2231 struct ptlrpc_request *close_req = NULL;
2232 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2235 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2236 OBD_ALLOC_PTR(op_data);
2237 if (op_data == NULL) {
2238 CWARN("%s: cannot allocate op_data to release open handle for "
2240 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2245 op_data->op_fid1 = body->mbo_fid1;
2246 op_data->op_handle = body->mbo_handle;
2247 op_data->op_mod_time = cfs_time_current_sec();
2248 md_close(exp, op_data, NULL, &close_req);
2249 ptlrpc_req_finished(close_req);
2250 ll_finish_md_op_data(op_data);
2255 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2256 struct super_block *sb, struct lookup_intent *it)
2258 struct ll_sb_info *sbi = NULL;
2259 struct lustre_md md = { NULL };
2263 LASSERT(*inode || sb);
2264 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2265 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2266 sbi->ll_md_exp, &md);
2271 rc = ll_update_inode(*inode, &md);
2275 LASSERT(sb != NULL);
2278 * At this point server returns to client's same fid as client
2279 * generated for creating. So using ->fid1 is okay here.
2281 if (!fid_is_sane(&md.body->mbo_fid1)) {
2282 CERROR("%s: Fid is insane "DFID"\n",
2283 ll_get_fsname(sb, NULL, 0),
2284 PFID(&md.body->mbo_fid1));
2285 GOTO(out, rc = -EINVAL);
2288 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2289 sbi->ll_flags & LL_SBI_32BIT_API),
2291 if (IS_ERR(*inode)) {
2292 #ifdef CONFIG_FS_POSIX_ACL
2294 posix_acl_release(md.posix_acl);
2295 md.posix_acl = NULL;
2298 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2300 CERROR("new_inode -fatal: rc %d\n", rc);
2305 /* Handling piggyback layout lock.
2306 * Layout lock can be piggybacked by getattr and open request.
2307 * The lsm can be applied to inode only if it comes with a layout lock
2308 * otherwise correct layout may be overwritten, for example:
2309 * 1. proc1: mdt returns a lsm but not granting layout
2310 * 2. layout was changed by another client
2311 * 3. proc2: refresh layout and layout lock granted
2312 * 4. proc1: to apply a stale layout */
2313 if (it != NULL && it->it_lock_mode != 0) {
2314 struct lustre_handle lockh;
2315 struct ldlm_lock *lock;
2317 lockh.cookie = it->it_lock_handle;
2318 lock = ldlm_handle2lock(&lockh);
2319 LASSERT(lock != NULL);
2320 if (ldlm_has_layout(lock)) {
2321 struct cl_object_conf conf;
2323 memset(&conf, 0, sizeof(conf));
2324 conf.coc_opc = OBJECT_CONF_SET;
2325 conf.coc_inode = *inode;
2326 conf.coc_lock = lock;
2327 conf.u.coc_layout = md.layout;
2328 (void)ll_layout_conf(*inode, &conf);
2330 LDLM_LOCK_PUT(lock);
2336 md_free_lustre_md(sbi->ll_md_exp, &md);
2339 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2340 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2345 int ll_obd_statfs(struct inode *inode, void __user *arg)
2347 struct ll_sb_info *sbi = NULL;
2348 struct obd_export *exp;
2350 struct obd_ioctl_data *data = NULL;
2354 if (!inode || !(sbi = ll_i2sbi(inode)))
2355 GOTO(out_statfs, rc = -EINVAL);
2357 rc = obd_ioctl_getdata(&buf, &len, arg);
2359 GOTO(out_statfs, rc);
2362 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2363 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2364 GOTO(out_statfs, rc = -EINVAL);
2366 if (data->ioc_inllen1 != sizeof(__u32) ||
2367 data->ioc_inllen2 != sizeof(__u32) ||
2368 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2369 data->ioc_plen2 != sizeof(struct obd_uuid))
2370 GOTO(out_statfs, rc = -EINVAL);
2372 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2373 if (type & LL_STATFS_LMV)
2374 exp = sbi->ll_md_exp;
2375 else if (type & LL_STATFS_LOV)
2376 exp = sbi->ll_dt_exp;
2378 GOTO(out_statfs, rc = -ENODEV);
2380 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2382 GOTO(out_statfs, rc);
2385 obd_ioctl_freedata(buf, len);
2389 int ll_process_config(struct lustre_cfg *lcfg)
2391 struct super_block *sb;
2396 /* The instance name contains the sb: lustre-client-aacfe000 */
2397 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2398 if (!ptr || !*(++ptr))
2400 if (sscanf(ptr, "%lx", &x) != 1)
2402 sb = (struct super_block *)x;
2403 /* This better be a real Lustre superblock! */
2404 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2406 /* Note we have not called client_common_fill_super yet, so
2407 proc fns must be able to handle that! */
2408 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2415 /* this function prepares md_op_data hint for passing it down to MD stack. */
2416 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2417 struct inode *i1, struct inode *i2,
2418 const char *name, size_t namelen,
2419 __u32 mode, __u32 opc, void *data)
2421 LASSERT(i1 != NULL);
2424 /* Do not reuse namelen for something else. */
2426 return ERR_PTR(-EINVAL);
2428 if (namelen > ll_i2sbi(i1)->ll_namelen)
2429 return ERR_PTR(-ENAMETOOLONG);
2431 if (!lu_name_is_valid_2(name, namelen))
2432 return ERR_PTR(-EINVAL);
2435 if (op_data == NULL)
2436 OBD_ALLOC_PTR(op_data);
2438 if (op_data == NULL)
2439 return ERR_PTR(-ENOMEM);
2441 ll_i2gids(op_data->op_suppgids, i1, i2);
2442 op_data->op_fid1 = *ll_inode2fid(i1);
2443 op_data->op_default_stripe_offset = -1;
2444 if (S_ISDIR(i1->i_mode)) {
2445 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2446 if (opc == LUSTRE_OPC_MKDIR)
2447 op_data->op_default_stripe_offset =
2448 ll_i2info(i1)->lli_def_stripe_offset;
2452 op_data->op_fid2 = *ll_inode2fid(i2);
2453 if (S_ISDIR(i2->i_mode))
2454 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2456 fid_zero(&op_data->op_fid2);
2459 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2460 op_data->op_cli_flags |= CLI_HASH64;
2462 if (ll_need_32bit_api(ll_i2sbi(i1)))
2463 op_data->op_cli_flags |= CLI_API32;
2465 op_data->op_name = name;
2466 op_data->op_namelen = namelen;
2467 op_data->op_mode = mode;
2468 op_data->op_mod_time = cfs_time_current_sec();
2469 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2470 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2471 op_data->op_cap = cfs_curproc_cap_pack();
2472 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2473 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2474 op_data->op_bias |= MDS_CREATE_VOLATILE;
2476 op_data->op_mds = 0;
2478 op_data->op_data = data;
2483 void ll_finish_md_op_data(struct md_op_data *op_data)
2485 security_release_secctx(op_data->op_file_secctx,
2486 op_data->op_file_secctx_size);
2487 OBD_FREE_PTR(op_data);
2490 #ifdef HAVE_SUPEROPS_USE_DENTRY
2491 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2493 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2496 struct ll_sb_info *sbi;
2498 #ifdef HAVE_SUPEROPS_USE_DENTRY
2499 LASSERT((seq != NULL) && (dentry != NULL));
2500 sbi = ll_s2sbi(dentry->d_sb);
2502 LASSERT((seq != NULL) && (vfs != NULL));
2503 sbi = ll_s2sbi(vfs->mnt_sb);
2506 if (sbi->ll_flags & LL_SBI_NOLCK)
2507 seq_puts(seq, ",nolock");
2509 if (sbi->ll_flags & LL_SBI_FLOCK)
2510 seq_puts(seq, ",flock");
2512 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2513 seq_puts(seq, ",localflock");
2515 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2516 seq_puts(seq, ",user_xattr");
2518 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2519 seq_puts(seq, ",lazystatfs");
2521 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2522 seq_puts(seq, ",user_fid2path");
2524 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2525 seq_puts(seq, ",always_ping");
2531 * Get obd name by cmd, and copy out to user space
2533 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2535 struct ll_sb_info *sbi = ll_i2sbi(inode);
2536 struct obd_device *obd;
2539 if (cmd == OBD_IOC_GETDTNAME)
2540 obd = class_exp2obd(sbi->ll_dt_exp);
2541 else if (cmd == OBD_IOC_GETMDNAME)
2542 obd = class_exp2obd(sbi->ll_md_exp);
2549 if (copy_to_user((void __user *)arg, obd->obd_name,
2550 strlen(obd->obd_name) + 1))
2557 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2558 * fsname will be returned in this buffer; otherwise, a static buffer will be
2559 * used to store the fsname and returned to caller.
2561 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2563 static char fsname_static[MTI_NAME_MAXLEN];
2564 struct lustre_sb_info *lsi = s2lsi(sb);
2569 /* this means the caller wants to use static buffer
2570 * and it doesn't care about race. Usually this is
2571 * in error reporting path */
2572 buf = fsname_static;
2573 buflen = sizeof(fsname_static);
2576 len = strlen(lsi->lsi_lmd->lmd_profile);
2577 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2578 if (ptr && (strcmp(ptr, "-client") == 0))
2581 if (unlikely(len >= buflen))
2583 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2589 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2596 p.mnt = current->fs->root.mnt;
2598 path = d_path(&p, buf, bufsize);
2603 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2605 char *buf, *path = NULL;
2606 struct dentry *dentry = NULL;
2607 struct inode *inode = page->mapping->host;
2609 /* this can be called inside spin lock so use GFP_ATOMIC. */
2610 buf = (char *)__get_free_page(GFP_ATOMIC);
2612 dentry = d_find_alias(page->mapping->host);
2614 path = ll_d_path(dentry, buf, PAGE_SIZE);
2618 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2619 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2620 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2621 PFID(ll_inode2fid(inode)),
2622 (path && !IS_ERR(path)) ? path : "", ioret);
2628 free_page((unsigned long)buf);
2631 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2632 struct lov_user_md **kbuf)
2634 struct lov_user_md lum;
2638 if (copy_from_user(&lum, md, sizeof(lum)))
2641 lum_size = ll_lov_user_md_size(&lum);
2645 OBD_ALLOC(*kbuf, lum_size);
2649 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2650 OBD_FREE(*kbuf, lum_size);
2658 * Compute llite root squash state after a change of root squash
2659 * configuration setting or add/remove of a lnet nid
2661 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2663 struct root_squash_info *squash = &sbi->ll_squash;
2666 lnet_process_id_t id;
2668 /* Update norootsquash flag */
2669 down_write(&squash->rsi_sem);
2670 if (list_empty(&squash->rsi_nosquash_nids))
2671 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2673 /* Do not apply root squash as soon as one of our NIDs is
2674 * in the nosquash_nids list */
2677 while (LNetGetId(i++, &id) != -ENOENT) {
2678 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2680 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2686 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2688 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2690 up_write(&squash->rsi_sem);
2694 * Parse linkea content to extract information about a given hardlink
2696 * \param[in] ldata - Initialized linkea data
2697 * \param[in] linkno - Link identifier
2698 * \param[out] parent_fid - The entry's parent FID
2699 * \param[out] ln - Entry name destination buffer
2701 * \retval 0 on success
2702 * \retval Appropriate negative error code on failure
2704 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2705 struct lu_fid *parent_fid, struct lu_name *ln)
2711 rc = linkea_init(ldata);
2715 if (linkno >= ldata->ld_leh->leh_reccount)
2716 /* beyond last link */
2719 linkea_first_entry(ldata);
2720 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2721 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2726 linkea_next_entry(ldata);
2736 * Get parent FID and name of an identified link. Operation is performed for
2737 * a given link number, letting the caller iterate over linkno to list one or
2738 * all links of an entry.
2740 * \param[in] file - File descriptor against which to perform the operation
2741 * \param[in,out] arg - User-filled structure containing the linkno to operate
2742 * on and the available size. It is eventually filled with
2743 * the requested information or left untouched on error
2745 * \retval - 0 on success
2746 * \retval - Appropriate negative error code on failure
2748 int ll_getparent(struct file *file, struct getparent __user *arg)
2750 struct dentry *dentry = file_dentry(file);
2751 struct inode *inode = file_inode(file);
2752 struct linkea_data *ldata;
2753 struct lu_buf buf = LU_BUF_NULL;
2755 struct lu_fid parent_fid;
2762 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2763 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2766 if (get_user(name_size, &arg->gp_name_size))
2769 if (get_user(linkno, &arg->gp_linkno))
2772 if (name_size > PATH_MAX)
2775 OBD_ALLOC(ldata, sizeof(*ldata));
2779 rc = linkea_data_new(ldata, &buf);
2781 GOTO(ldata_free, rc);
2783 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2787 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2791 if (ln.ln_namelen >= name_size)
2792 GOTO(lb_free, rc = -EOVERFLOW);
2794 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2795 GOTO(lb_free, rc = -EFAULT);
2797 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2798 GOTO(lb_free, rc = -EFAULT);
2800 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2801 GOTO(lb_free, rc = -EFAULT);
2806 OBD_FREE(ldata, sizeof(*ldata));