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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 cfs_mem_cache_t *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
97 lru_page_max = pages / 2;
99 lru_page_max = (pages / 4) * 3;
102 /* initialize ll_cache data */
103 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
104 sbi->ll_cache.ccc_lru_max = lru_page_max;
105 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
106 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
107 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
109 cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
110 cfs_waitq_init(&sbi->ll_cache.ccc_unstable_waitq);
112 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
113 SBI_DEFAULT_READAHEAD_MAX);
114 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
115 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
116 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
117 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
118 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
120 ll_generate_random_uuid(uuid);
121 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
122 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
124 spin_lock(&ll_sb_lock);
125 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
126 spin_unlock(&ll_sb_lock);
128 sbi->ll_flags |= LL_SBI_VERBOSE;
129 #ifdef ENABLE_CHECKSUM
130 sbi->ll_flags |= LL_SBI_CHECKSUM;
133 #ifdef HAVE_LRU_RESIZE_SUPPORT
134 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
137 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
138 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
140 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
144 /* metadata statahead is enabled by default */
145 sbi->ll_sa_max = LL_SA_RPC_DEF;
146 cfs_atomic_set(&sbi->ll_sa_total, 0);
147 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
148 cfs_atomic_set(&sbi->ll_agl_total, 0);
149 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
154 void ll_free_sbi(struct super_block *sb)
156 struct ll_sb_info *sbi = ll_s2sbi(sb);
160 spin_lock(&ll_sb_lock);
161 cfs_list_del(&sbi->ll_list);
162 spin_unlock(&ll_sb_lock);
163 OBD_FREE(sbi, sizeof(*sbi));
168 static struct dentry_operations ll_d_root_ops = {
169 .d_compare = ll_dcompare,
170 .d_revalidate = ll_revalidate_nd,
173 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
174 struct vfsmount *mnt)
176 struct inode *root = 0;
177 struct ll_sb_info *sbi = ll_s2sbi(sb);
178 struct obd_device *obd;
179 struct obd_capa *oc = NULL;
180 struct obd_statfs *osfs = NULL;
181 struct ptlrpc_request *request = NULL;
182 struct obd_connect_data *data = NULL;
183 struct obd_uuid *uuid;
184 struct md_op_data *op_data;
185 struct lustre_md lmd;
187 int size, err, checksum;
190 obd = class_name2obd(md);
192 CERROR("MD %s: not setup or attached\n", md);
206 if (proc_lustre_fs_root) {
207 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
210 CERROR("could not register mount in /proc/fs/lustre\n");
213 /* indicate the features supported by this client */
214 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
215 OBD_CONNECT_ATTRFID |
216 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
217 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
218 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
219 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
220 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
221 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
222 OBD_CONNECT_EINPROGRESS |
223 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
224 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
226 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
227 data->ocd_connect_flags |= OBD_CONNECT_SOM;
229 #ifdef HAVE_LRU_RESIZE_SUPPORT
230 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
231 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
233 #ifdef CONFIG_FS_POSIX_ACL
234 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
237 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
238 /* flag mdc connection as lightweight, only used for test
239 * purpose, use with care */
240 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
242 data->ocd_ibits_known = MDS_INODELOCK_FULL;
243 data->ocd_version = LUSTRE_VERSION_CODE;
245 if (sb->s_flags & MS_RDONLY)
246 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
247 if (sbi->ll_flags & LL_SBI_USER_XATTR)
248 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
250 #ifdef HAVE_MS_FLOCK_LOCK
251 /* force vfs to use lustre handler for flock() calls - bug 10743 */
252 sb->s_flags |= MS_FLOCK_LOCK;
254 #ifdef MS_HAS_NEW_AOPS
255 sb->s_flags |= MS_HAS_NEW_AOPS;
258 if (sbi->ll_flags & LL_SBI_FLOCK)
259 sbi->ll_fop = &ll_file_operations_flock;
260 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
261 sbi->ll_fop = &ll_file_operations;
263 sbi->ll_fop = &ll_file_operations_noflock;
266 data->ocd_connect_flags |= OBD_CONNECT_REAL;
267 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
268 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
270 data->ocd_brw_size = MD_MAX_BRW_SIZE;
272 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
274 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
275 "recovery, of which this client is not a "
276 "part. Please wait for recovery to complete,"
277 " abort, or time out.\n", md);
280 CERROR("cannot connect to %s: rc = %d\n", md, err);
284 sbi->ll_md_exp->exp_connect_data = *data;
286 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
287 LUSTRE_SEQ_METADATA);
289 CERROR("%s: Can't init metadata layer FID infrastructure, "
290 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
294 /* For mount, we only need fs info from MDT0, and also in DNE, it
295 * can make sure the client can be mounted as long as MDT0 is
297 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
298 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
299 OBD_STATFS_FOR_MDT0);
301 GOTO(out_md_fid, err);
303 /* This needs to be after statfs to ensure connect has finished.
304 * Note that "data" does NOT contain the valid connect reply.
305 * If connecting to a 1.8 server there will be no LMV device, so
306 * we can access the MDC export directly and exp_connect_flags will
307 * be non-zero, but if accessing an upgraded 2.1 server it will
308 * have the correct flags filled in.
309 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
310 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
311 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
312 valid != CLIENT_CONNECT_MDT_REQD) {
315 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
316 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
317 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
318 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
319 "feature(s) needed for correct operation "
320 "of this client (%s). Please upgrade "
321 "server or downgrade client.\n",
322 sbi->ll_md_exp->exp_obd->obd_name, buf);
323 OBD_FREE(buf, CFS_PAGE_SIZE);
324 GOTO(out_md_fid, err = -EPROTO);
327 size = sizeof(*data);
328 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
329 KEY_CONN_DATA, &size, data, NULL);
331 CERROR("%s: Get connect data failed: rc = %d\n",
332 sbi->ll_md_exp->exp_obd->obd_name, err);
333 GOTO(out_md_fid, err);
336 LASSERT(osfs->os_bsize);
337 sb->s_blocksize = osfs->os_bsize;
338 sb->s_blocksize_bits = log2(osfs->os_bsize);
339 sb->s_magic = LL_SUPER_MAGIC;
340 sb->s_maxbytes = MAX_LFS_FILESIZE;
341 sbi->ll_namelen = osfs->os_namelen;
342 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
344 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
345 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
346 LCONSOLE_INFO("Disabling user_xattr feature because "
347 "it is not supported on the server\n");
348 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
351 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
353 sb->s_flags |= MS_POSIXACL;
355 sbi->ll_flags |= LL_SBI_ACL;
357 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
359 sb->s_flags &= ~MS_POSIXACL;
361 sbi->ll_flags &= ~LL_SBI_ACL;
364 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
365 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
366 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
367 LCONSOLE_INFO("client is set as remote by default.\n");
370 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
371 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
372 LCONSOLE_INFO("client claims to be remote, but server "
373 "rejected, forced to be local.\n");
377 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
378 LCONSOLE_INFO("client enabled MDS capability!\n");
379 sbi->ll_flags |= LL_SBI_MDS_CAPA;
382 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
383 LCONSOLE_INFO("client enabled OSS capability!\n");
384 sbi->ll_flags |= LL_SBI_OSS_CAPA;
387 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
388 sbi->ll_flags |= LL_SBI_64BIT_HASH;
390 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
391 sbi->ll_md_brw_size = data->ocd_brw_size;
393 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
395 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
396 LCONSOLE_INFO("Layout lock feature supported.\n");
397 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
400 obd = class_name2obd(dt);
402 CERROR("DT %s: not setup or attached\n", dt);
403 GOTO(out_md_fid, err = -ENODEV);
406 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
407 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
408 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
409 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
410 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
411 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
412 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
413 OBD_CONNECT_MAXBYTES |
414 OBD_CONNECT_EINPROGRESS |
415 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
416 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
418 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
419 data->ocd_connect_flags |= OBD_CONNECT_SOM;
421 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
422 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
423 * disabled by default, because it can still be enabled on the
424 * fly via /proc. As a consequence, we still need to come to an
425 * agreement on the supported algorithms at connect time */
426 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
428 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
429 data->ocd_cksum_types = OBD_CKSUM_ADLER;
431 data->ocd_cksum_types = cksum_types_supported_client();
434 #ifdef HAVE_LRU_RESIZE_SUPPORT
435 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
437 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
438 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
440 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
441 "ocd_grant: %d\n", data->ocd_connect_flags,
442 data->ocd_version, data->ocd_grant);
444 obd->obd_upcall.onu_owner = &sbi->ll_lco;
445 obd->obd_upcall.onu_upcall = cl_ocd_update;
447 data->ocd_brw_size = DT_MAX_BRW_SIZE;
449 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
452 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
453 "recovery, of which this client is not a "
454 "part. Please wait for recovery to "
455 "complete, abort, or time out.\n", dt);
458 CERROR("%s: Cannot connect to %s: rc = %d\n",
459 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
463 sbi->ll_dt_exp->exp_connect_data = *data;
465 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
466 LUSTRE_SEQ_METADATA);
468 CERROR("%s: Can't init data layer FID infrastructure, "
469 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
473 mutex_lock(&sbi->ll_lco.lco_lock);
474 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
475 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
476 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
477 mutex_unlock(&sbi->ll_lco.lco_lock);
479 fid_zero(&sbi->ll_root_fid);
480 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
482 CERROR("cannot mds_connect: rc = %d\n", err);
483 GOTO(out_lock_cn_cb, err);
485 if (!fid_is_sane(&sbi->ll_root_fid)) {
486 CERROR("%s: Invalid root fid "DFID" during mount\n",
487 sbi->ll_md_exp->exp_obd->obd_name,
488 PFID(&sbi->ll_root_fid));
489 GOTO(out_lock_cn_cb, err = -EINVAL);
491 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
493 sb->s_op = &lustre_super_operations;
494 #if THREAD_SIZE >= 8192 /*b=17630*/
495 sb->s_export_op = &lustre_export_operations;
499 * XXX: move this to after cbd setup? */
500 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
501 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
502 valid |= OBD_MD_FLRMTPERM;
503 else if (sbi->ll_flags & LL_SBI_ACL)
504 valid |= OBD_MD_FLACL;
506 OBD_ALLOC_PTR(op_data);
508 GOTO(out_lock_cn_cb, err = -ENOMEM);
510 op_data->op_fid1 = sbi->ll_root_fid;
511 op_data->op_mode = 0;
512 op_data->op_capa1 = oc;
513 op_data->op_valid = valid;
515 err = md_getattr(sbi->ll_md_exp, op_data, &request);
518 OBD_FREE_PTR(op_data);
520 CERROR("%s: md_getattr failed for root: rc = %d\n",
521 sbi->ll_md_exp->exp_obd->obd_name, err);
522 GOTO(out_lock_cn_cb, err);
525 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
526 sbi->ll_md_exp, &lmd);
528 CERROR("failed to understand root inode md: rc = %d\n", err);
529 ptlrpc_req_finished(request);
530 GOTO(out_lock_cn_cb, err);
533 LASSERT(fid_is_sane(&sbi->ll_root_fid));
534 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
535 ll_need_32bit_api(sbi)),
537 md_free_lustre_md(sbi->ll_md_exp, &lmd);
538 ptlrpc_req_finished(request);
540 if (root == NULL || IS_ERR(root)) {
542 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
543 #ifdef CONFIG_FS_POSIX_ACL
545 posix_acl_release(lmd.posix_acl);
546 lmd.posix_acl = NULL;
549 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
551 CERROR("lustre_lite: bad iget4 for root\n");
555 err = ll_close_thread_start(&sbi->ll_lcq);
557 CERROR("cannot start close thread: rc %d\n", err);
561 #ifdef CONFIG_FS_POSIX_ACL
562 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
563 rct_init(&sbi->ll_rct);
564 et_init(&sbi->ll_et);
568 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
569 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
570 KEY_CHECKSUM, sizeof(checksum), &checksum,
574 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
575 KEY_CACHE_SET, sizeof(sbi->ll_cache),
576 &sbi->ll_cache, NULL);
578 sb->s_root = d_make_root(root);
579 if (sb->s_root == NULL) {
580 CERROR("%s: can't make root dentry\n",
581 ll_get_fsname(sb, NULL, 0));
582 GOTO(out_root, err = -ENOMEM);
585 #ifdef HAVE_DCACHE_LOCK
586 sb->s_root->d_op = &ll_d_root_ops;
588 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
589 d_set_d_op(sb->s_root, &ll_d_root_ops);
590 sb->s_d_op = &ll_d_ops;
593 sbi->ll_sdev_orig = sb->s_dev;
595 /* We set sb->s_dev equal on all lustre clients in order to support
596 * NFS export clustering. NFSD requires that the FSID be the same
598 /* s_dev is also used in lt_compare() to compare two fs, but that is
599 * only a node-local comparison. */
600 uuid = obd_get_uuid(sbi->ll_md_exp);
602 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
614 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
616 obd_disconnect(sbi->ll_dt_exp);
617 sbi->ll_dt_exp = NULL;
618 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
619 obd_zombie_barrier();
621 obd_fid_fini(sbi->ll_md_exp->exp_obd);
623 obd_disconnect(sbi->ll_md_exp);
624 sbi->ll_md_exp = NULL;
630 lprocfs_unregister_mountpoint(sbi);
634 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
638 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
640 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
641 KEY_MAX_EASIZE, &size, lmmsize, NULL);
643 CERROR("Get max mdsize error rc %d \n", rc);
648 void ll_dump_inode(struct inode *inode)
650 struct ll_d_hlist_node *tmp;
651 int dentry_count = 0;
653 LASSERT(inode != NULL);
655 ll_d_hlist_for_each(tmp, &inode->i_dentry)
658 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
659 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
660 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
663 void lustre_dump_dentry(struct dentry *dentry, int recur)
665 struct list_head *tmp;
668 LASSERT(dentry != NULL);
670 list_for_each(tmp, &dentry->d_subdirs)
673 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
674 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
675 dentry->d_name.len, dentry->d_name.name,
676 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
677 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
678 dentry->d_flags, dentry->d_fsdata, subdirs);
679 if (dentry->d_inode != NULL)
680 ll_dump_inode(dentry->d_inode);
685 list_for_each(tmp, &dentry->d_subdirs) {
686 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
687 lustre_dump_dentry(d, recur - 1);
691 void client_common_put_super(struct super_block *sb)
693 struct ll_sb_info *sbi = ll_s2sbi(sb);
696 #ifdef CONFIG_FS_POSIX_ACL
697 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
698 et_fini(&sbi->ll_et);
699 rct_fini(&sbi->ll_rct);
703 ll_close_thread_shutdown(sbi->ll_lcq);
707 cfs_list_del(&sbi->ll_conn_chain);
709 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
710 obd_disconnect(sbi->ll_dt_exp);
711 sbi->ll_dt_exp = NULL;
712 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
714 obd_zombie_barrier();
716 lprocfs_unregister_mountpoint(sbi);
718 obd_fid_fini(sbi->ll_md_exp->exp_obd);
719 obd_disconnect(sbi->ll_md_exp);
720 sbi->ll_md_exp = NULL;
725 void ll_kill_super(struct super_block *sb)
727 struct ll_sb_info *sbi;
732 if (!(sb->s_flags & MS_ACTIVE))
736 /* we need restore s_dev from changed for clustred NFS before put_super
737 * because new kernels have cached s_dev and change sb->s_dev in
738 * put_super not affected real removing devices */
740 sb->s_dev = sbi->ll_sdev_orig;
744 char *ll_read_opt(const char *opt, char *data)
750 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
751 if (strncmp(opt, data, strlen(opt)))
753 if ((value = strchr(data, '=')) == NULL)
757 OBD_ALLOC(retval, strlen(value) + 1);
759 CERROR("out of memory!\n");
763 memcpy(retval, value, strlen(value)+1);
764 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
768 static inline int ll_set_opt(const char *opt, char *data, int fl)
770 if (strncmp(opt, data, strlen(opt)) != 0)
776 /* non-client-specific mount options are parsed in lmd_parse */
777 static int ll_options(char *options, int *flags)
780 char *s1 = options, *s2;
786 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
789 CDEBUG(D_SUPER, "next opt=%s\n", s1);
790 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
795 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
800 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
805 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
810 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
815 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
820 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
821 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
823 /* Ignore deprecated mount option. The client will
824 * always try to mount with ACL support, whether this
825 * is used depends on whether server supports it. */
826 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
827 "mount option 'acl'.\n");
830 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
832 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
833 "mount option 'noacl'.\n");
837 #warning "{no}acl options have been deprecated since 1.8, please remove them"
839 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
844 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
849 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
855 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
860 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
865 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
870 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
875 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
880 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
885 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
890 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
895 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
900 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
905 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
911 s2 = strchr(s1, ',');
919 void ll_lli_init(struct ll_inode_info *lli)
921 lli->lli_inode_magic = LLI_INODE_MAGIC;
923 lli->lli_ioepoch = 0;
924 lli->lli_maxbytes = MAX_LFS_FILESIZE;
925 spin_lock_init(&lli->lli_lock);
926 lli->lli_posix_acl = NULL;
927 lli->lli_remote_perms = NULL;
928 mutex_init(&lli->lli_rmtperm_mutex);
929 /* Do not set lli_fid, it has been initialized already. */
930 fid_zero(&lli->lli_pfid);
931 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
932 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
933 cfs_atomic_set(&lli->lli_open_count, 0);
934 lli->lli_mds_capa = NULL;
935 lli->lli_rmtperm_time = 0;
936 lli->lli_pending_och = NULL;
937 lli->lli_mds_read_och = NULL;
938 lli->lli_mds_write_och = NULL;
939 lli->lli_mds_exec_och = NULL;
940 lli->lli_open_fd_read_count = 0;
941 lli->lli_open_fd_write_count = 0;
942 lli->lli_open_fd_exec_count = 0;
943 mutex_init(&lli->lli_och_mutex);
944 spin_lock_init(&lli->lli_agl_lock);
945 lli->lli_has_smd = false;
946 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
947 lli->lli_clob = NULL;
949 LASSERT(lli->lli_vfs_inode.i_mode != 0);
950 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
951 mutex_init(&lli->lli_readdir_mutex);
952 lli->lli_opendir_key = NULL;
954 lli->lli_def_acl = NULL;
955 spin_lock_init(&lli->lli_sa_lock);
956 lli->lli_opendir_pid = 0;
958 sema_init(&lli->lli_size_sem, 1);
959 lli->lli_size_sem_owner = NULL;
960 lli->lli_symlink_name = NULL;
961 init_rwsem(&lli->lli_trunc_sem);
962 mutex_init(&lli->lli_write_mutex);
963 init_rwsem(&lli->lli_glimpse_sem);
964 lli->lli_glimpse_time = 0;
965 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
966 lli->lli_agl_index = 0;
967 lli->lli_async_rc = 0;
968 lli->lli_volatile = false;
970 mutex_init(&lli->lli_layout_mutex);
973 static inline int ll_bdi_register(struct backing_dev_info *bdi)
975 #ifdef HAVE_BDI_REGISTER
976 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
979 bdi->name = "lustre";
981 return bdi_register(bdi, NULL, "lustre-%d",
982 atomic_inc_return(&ll_bdi_num));
988 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
990 struct lustre_profile *lprof = NULL;
991 struct lustre_sb_info *lsi = s2lsi(sb);
992 struct ll_sb_info *sbi;
993 char *dt = NULL, *md = NULL;
994 char *profilenm = get_profile_name(sb);
995 struct config_llog_instance *cfg;
996 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
997 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1001 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1009 /* client additional sb info */
1010 lsi->lsi_llsbi = sbi = ll_init_sbi();
1012 cfs_module_put(THIS_MODULE);
1017 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1019 GOTO(out_free, err);
1021 err = bdi_init(&lsi->lsi_bdi);
1023 GOTO(out_free, err);
1024 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1025 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1026 err = ll_bdi_register(&lsi->lsi_bdi);
1028 GOTO(out_free, err);
1031 sb->s_bdi = &lsi->lsi_bdi;
1034 /* Generate a string unique to this super, in case some joker tries
1035 to mount the same fs at two mount points.
1036 Use the address of the super itself.*/
1037 cfg->cfg_instance = sb;
1038 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1039 cfg->cfg_callback = class_config_llog_handler;
1040 /* set up client obds */
1041 err = lustre_process_log(sb, profilenm, cfg);
1043 CERROR("Unable to process log: %d\n", err);
1044 GOTO(out_free, err);
1047 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1048 lprof = class_get_profile(profilenm);
1049 if (lprof == NULL) {
1050 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1051 " read from the MGS. Does that filesystem "
1052 "exist?\n", profilenm);
1053 GOTO(out_free, err = -EINVAL);
1055 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1056 lprof->lp_md, lprof->lp_dt);
1058 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1060 GOTO(out_free, err = -ENOMEM);
1061 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1063 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1065 GOTO(out_free, err = -ENOMEM);
1066 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1068 /* connections, registrations, sb setup */
1069 err = client_common_fill_super(sb, md, dt, mnt);
1073 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1075 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1078 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1079 LCONSOLE_WARN("Mounted %s\n", profilenm);
1083 } /* ll_fill_super */
1086 void lu_context_keys_dump(void);
1088 void ll_put_super(struct super_block *sb)
1090 struct config_llog_instance cfg;
1091 struct obd_device *obd;
1092 struct lustre_sb_info *lsi = s2lsi(sb);
1093 struct ll_sb_info *sbi = ll_s2sbi(sb);
1094 char *profilenm = get_profile_name(sb);
1095 int ccc_count, next, force = 1, rc = 0;
1098 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1100 ll_print_capa_stat(sbi);
1102 cfg.cfg_instance = sb;
1103 lustre_end_log(sb, profilenm, &cfg);
1105 if (sbi->ll_md_exp) {
1106 obd = class_exp2obd(sbi->ll_md_exp);
1108 force = obd->obd_force;
1111 /* Wait for unstable pages to be committed to stable storage */
1113 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1114 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1115 cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1119 ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1120 if (force == 0 && rc != -EINTR)
1121 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1124 /* We need to set force before the lov_disconnect in
1125 lustre_common_put_super, since l_d cleans up osc's as well. */
1128 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1130 obd->obd_force = force;
1135 /* Only if client_common_fill_super succeeded */
1136 client_common_put_super(sb);
1140 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1141 class_manual_cleanup(obd);
1144 if (sbi->ll_flags & LL_SBI_VERBOSE)
1145 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1148 class_del_profile(profilenm);
1150 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1151 bdi_destroy(&lsi->lsi_bdi);
1152 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1156 lsi->lsi_llsbi = NULL;
1158 lustre_common_put_super(sb);
1160 cfs_module_put(THIS_MODULE);
1163 } /* client_put_super */
1165 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1167 struct inode *inode = NULL;
1169 /* NOTE: we depend on atomic igrab() -bzzz */
1170 lock_res_and_lock(lock);
1171 if (lock->l_resource->lr_lvb_inode) {
1172 struct ll_inode_info * lli;
1173 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1174 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1175 inode = igrab(lock->l_resource->lr_lvb_inode);
1177 inode = lock->l_resource->lr_lvb_inode;
1178 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1179 D_WARNING, lock, "lr_lvb_inode %p is "
1180 "bogus: magic %08x",
1181 lock->l_resource->lr_lvb_inode,
1182 lli->lli_inode_magic);
1186 unlock_res_and_lock(lock);
1190 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1192 struct inode *inode = NULL;
1193 /* NOTE: we depend on atomic igrab() -bzzz */
1194 lock_res_and_lock(lock);
1195 if (lock->l_ast_data) {
1196 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1197 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1198 inode = igrab(lock->l_ast_data);
1200 inode = lock->l_ast_data;
1201 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1202 D_WARNING, lock, "l_ast_data %p is "
1203 "bogus: magic %08x", lock->l_ast_data,
1204 lli->lli_inode_magic);
1208 unlock_res_and_lock(lock);
1212 void ll_clear_inode(struct inode *inode)
1214 struct ll_inode_info *lli = ll_i2info(inode);
1215 struct ll_sb_info *sbi = ll_i2sbi(inode);
1218 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1219 inode->i_generation, inode);
1221 if (S_ISDIR(inode->i_mode)) {
1222 /* these should have been cleared in ll_file_release */
1223 LASSERT(lli->lli_opendir_key == NULL);
1224 LASSERT(lli->lli_sai == NULL);
1225 LASSERT(lli->lli_opendir_pid == 0);
1228 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1229 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1231 LASSERT(!lli->lli_open_fd_write_count);
1232 LASSERT(!lli->lli_open_fd_read_count);
1233 LASSERT(!lli->lli_open_fd_exec_count);
1235 if (lli->lli_mds_write_och)
1236 ll_md_real_close(inode, FMODE_WRITE);
1237 if (lli->lli_mds_exec_och)
1238 ll_md_real_close(inode, FMODE_EXEC);
1239 if (lli->lli_mds_read_och)
1240 ll_md_real_close(inode, FMODE_READ);
1242 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1243 OBD_FREE(lli->lli_symlink_name,
1244 strlen(lli->lli_symlink_name) + 1);
1245 lli->lli_symlink_name = NULL;
1248 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1249 LASSERT(lli->lli_posix_acl == NULL);
1250 if (lli->lli_remote_perms) {
1251 free_rmtperm_hash(lli->lli_remote_perms);
1252 lli->lli_remote_perms = NULL;
1255 #ifdef CONFIG_FS_POSIX_ACL
1256 else if (lli->lli_posix_acl) {
1257 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1258 LASSERT(lli->lli_remote_perms == NULL);
1259 posix_acl_release(lli->lli_posix_acl);
1260 lli->lli_posix_acl = NULL;
1263 lli->lli_inode_magic = LLI_INODE_DEAD;
1265 ll_clear_inode_capas(inode);
1266 if (!S_ISDIR(inode->i_mode))
1267 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1270 * XXX This has to be done before lsm is freed below, because
1271 * cl_object still uses inode lsm.
1273 cl_inode_fini(inode);
1274 lli->lli_has_smd = false;
1279 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1280 struct md_open_data **mod)
1282 struct lustre_md md;
1283 struct inode *inode = dentry->d_inode;
1284 struct ll_sb_info *sbi = ll_i2sbi(inode);
1285 struct ptlrpc_request *request = NULL;
1289 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1290 LUSTRE_OPC_ANY, NULL);
1291 if (IS_ERR(op_data))
1292 RETURN(PTR_ERR(op_data));
1294 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1297 ptlrpc_req_finished(request);
1298 if (rc == -ENOENT) {
1300 /* Unlinked special device node? Or just a race?
1301 * Pretend we done everything. */
1302 if (!S_ISREG(inode->i_mode) &&
1303 !S_ISDIR(inode->i_mode)) {
1304 ia_valid = op_data->op_attr.ia_valid;
1305 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1306 rc = simple_setattr(dentry, &op_data->op_attr);
1307 op_data->op_attr.ia_valid = ia_valid;
1309 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1310 CERROR("md_setattr fails: rc = %d\n", rc);
1315 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1316 sbi->ll_md_exp, &md);
1318 ptlrpc_req_finished(request);
1322 ia_valid = op_data->op_attr.ia_valid;
1323 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1324 * cache is not cleared yet. */
1325 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1326 rc = simple_setattr(dentry, &op_data->op_attr);
1327 op_data->op_attr.ia_valid = ia_valid;
1329 /* Extract epoch data if obtained. */
1330 op_data->op_handle = md.body->handle;
1331 op_data->op_ioepoch = md.body->ioepoch;
1333 ll_update_inode(inode, &md);
1334 ptlrpc_req_finished(request);
1339 /* Close IO epoch and send Size-on-MDS attribute update. */
1340 static int ll_setattr_done_writing(struct inode *inode,
1341 struct md_op_data *op_data,
1342 struct md_open_data *mod)
1344 struct ll_inode_info *lli = ll_i2info(inode);
1348 LASSERT(op_data != NULL);
1349 if (!S_ISREG(inode->i_mode))
1352 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1353 op_data->op_ioepoch, PFID(&lli->lli_fid));
1355 op_data->op_flags = MF_EPOCH_CLOSE;
1356 ll_done_writing_attr(inode, op_data);
1357 ll_pack_inode2opdata(inode, op_data, NULL);
1359 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1360 if (rc == -EAGAIN) {
1361 /* MDS has instructed us to obtain Size-on-MDS attribute
1362 * from OSTs and send setattr to back to MDS. */
1363 rc = ll_som_update(inode, op_data);
1365 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1371 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1373 struct obd_capa *capa;
1376 if (attr->ia_valid & ATTR_SIZE)
1377 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1379 capa = ll_mdscapa_get(inode);
1381 rc = cl_setattr_ost(inode, attr, capa);
1383 if (attr->ia_valid & ATTR_SIZE)
1384 ll_truncate_free_capa(capa);
1391 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1393 * inode_newsize_ok - may this inode be truncated to a given size
1394 * @inode: the inode to be truncated
1395 * @offset: the new size to assign to the inode
1396 * @Returns: 0 on success, -ve errno on failure
1398 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1399 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1400 * when necessary. Caller must not proceed with inode size change if failure is
1401 * returned. @inode must be a file (not directory), with appropriate
1402 * permissions to allow truncate (inode_newsize_ok does NOT check these
1405 * inode_newsize_ok must be called with i_mutex held.
1407 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1409 if (inode->i_size < offset) {
1410 unsigned long limit;
1412 limit = rlimit(RLIMIT_FSIZE);
1413 if (limit != RLIM_INFINITY && offset > limit)
1415 if (offset > inode->i_sb->s_maxbytes)
1419 * truncation of in-use swapfiles is disallowed - it would
1420 * cause subsequent swapout to scribble on the now-freed
1423 if (IS_SWAPFILE(inode))
1429 send_sig(SIGXFSZ, current, 0);
1435 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1436 * object(s) determine the file size and mtime. Otherwise, the MDS will
1437 * keep these values until such a time that objects are allocated for it.
1438 * We do the MDS operations first, as it is checking permissions for us.
1439 * We don't to the MDS RPC if there is nothing that we want to store there,
1440 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1441 * going to do an RPC anyways.
1443 * If we are doing a truncate, we will send the mtime and ctime updates
1444 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1445 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1448 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1450 struct inode *inode = dentry->d_inode;
1451 struct ll_inode_info *lli = ll_i2info(inode);
1452 struct md_op_data *op_data = NULL;
1453 struct md_open_data *mod = NULL;
1454 int rc = 0, rc1 = 0;
1457 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1458 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1459 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1462 if (attr->ia_valid & ATTR_SIZE) {
1463 /* Check new size against VFS/VM file size limit and rlimit */
1464 rc = inode_newsize_ok(inode, attr->ia_size);
1468 /* The maximum Lustre file size is variable, based on the
1469 * OST maximum object size and number of stripes. This
1470 * needs another check in addition to the VFS check above. */
1471 if (attr->ia_size > ll_file_maxbytes(inode)) {
1472 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1473 PFID(&lli->lli_fid), attr->ia_size,
1474 ll_file_maxbytes(inode));
1478 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1481 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1482 if (attr->ia_valid & TIMES_SET_FLAGS) {
1483 if (cfs_curproc_fsuid() != inode->i_uid &&
1484 !cfs_capable(CFS_CAP_FOWNER))
1488 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1489 if (attr->ia_valid & ATTR_CTIME) {
1490 attr->ia_ctime = CFS_CURRENT_TIME;
1491 attr->ia_valid |= ATTR_CTIME_SET;
1493 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1494 (attr->ia_valid & ATTR_ATIME)) {
1495 attr->ia_atime = CFS_CURRENT_TIME;
1496 attr->ia_valid |= ATTR_ATIME_SET;
1498 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1499 (attr->ia_valid & ATTR_MTIME)) {
1500 attr->ia_mtime = CFS_CURRENT_TIME;
1501 attr->ia_valid |= ATTR_MTIME_SET;
1504 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1505 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1506 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1507 cfs_time_current_sec());
1509 /* If we are changing file size, file content is modified, flag it. */
1510 if (attr->ia_valid & ATTR_SIZE) {
1511 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1512 spin_lock(&lli->lli_lock);
1513 lli->lli_flags |= LLIF_DATA_MODIFIED;
1514 spin_unlock(&lli->lli_lock);
1517 /* We always do an MDS RPC, even if we're only changing the size;
1518 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1520 OBD_ALLOC_PTR(op_data);
1521 if (op_data == NULL)
1524 if (!S_ISDIR(inode->i_mode)) {
1525 if (attr->ia_valid & ATTR_SIZE)
1526 inode_dio_write_done(inode);
1527 mutex_unlock(&inode->i_mutex);
1528 down_write(&lli->lli_trunc_sem);
1531 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1533 /* Open epoch for truncate. */
1534 if (exp_connect_som(ll_i2mdexp(inode)) &&
1535 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1536 op_data->op_flags = MF_EPOCH_OPEN;
1538 rc = ll_md_setattr(dentry, op_data, &mod);
1542 /* RPC to MDT is sent, cancel data modification flag */
1543 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1544 spin_lock(&lli->lli_lock);
1545 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1546 spin_unlock(&lli->lli_lock);
1549 ll_ioepoch_open(lli, op_data->op_ioepoch);
1550 if (!S_ISREG(inode->i_mode))
1553 if (attr->ia_valid & (ATTR_SIZE |
1554 ATTR_ATIME | ATTR_ATIME_SET |
1555 ATTR_MTIME | ATTR_MTIME_SET))
1556 /* For truncate and utimes sending attributes to OSTs, setting
1557 * mtime/atime to the past will be performed under PW [0:EOF]
1558 * extent lock (new_size:EOF for truncate). It may seem
1559 * excessive to send mtime/atime updates to OSTs when not
1560 * setting times to past, but it is necessary due to possible
1561 * time de-synchronization between MDT inode and OST objects */
1562 rc = ll_setattr_ost(inode, attr);
1566 if (op_data->op_ioepoch) {
1567 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1571 ll_finish_md_op_data(op_data);
1573 if (!S_ISDIR(inode->i_mode)) {
1574 up_write(&lli->lli_trunc_sem);
1575 mutex_lock(&inode->i_mutex);
1576 if (attr->ia_valid & ATTR_SIZE)
1577 inode_dio_wait(inode);
1580 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1581 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1586 int ll_setattr(struct dentry *de, struct iattr *attr)
1588 int mode = de->d_inode->i_mode;
1590 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1591 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1592 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1594 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1595 (ATTR_SIZE|ATTR_MODE)) &&
1596 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1597 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1598 !(attr->ia_mode & S_ISGID))))
1599 attr->ia_valid |= ATTR_FORCE;
1601 if ((mode & S_ISUID) &&
1602 !(attr->ia_mode & S_ISUID) &&
1603 !(attr->ia_valid & ATTR_KILL_SUID))
1604 attr->ia_valid |= ATTR_KILL_SUID;
1606 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1607 !(attr->ia_mode & S_ISGID) &&
1608 !(attr->ia_valid & ATTR_KILL_SGID))
1609 attr->ia_valid |= ATTR_KILL_SGID;
1611 return ll_setattr_raw(de, attr);
1614 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1615 __u64 max_age, __u32 flags)
1617 struct ll_sb_info *sbi = ll_s2sbi(sb);
1618 struct obd_statfs obd_osfs;
1622 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1624 CERROR("md_statfs fails: rc = %d\n", rc);
1628 osfs->os_type = sb->s_magic;
1630 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1631 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1633 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1634 flags |= OBD_STATFS_NODELAY;
1636 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1638 CERROR("obd_statfs fails: rc = %d\n", rc);
1642 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1643 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1646 osfs->os_bsize = obd_osfs.os_bsize;
1647 osfs->os_blocks = obd_osfs.os_blocks;
1648 osfs->os_bfree = obd_osfs.os_bfree;
1649 osfs->os_bavail = obd_osfs.os_bavail;
1651 /* If we don't have as many objects free on the OST as inodes
1652 * on the MDS, we reduce the total number of inodes to
1653 * compensate, so that the "inodes in use" number is correct.
1655 if (obd_osfs.os_ffree < osfs->os_ffree) {
1656 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1658 osfs->os_ffree = obd_osfs.os_ffree;
1663 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1665 struct super_block *sb = de->d_sb;
1666 struct obd_statfs osfs;
1669 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1670 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1672 /* Some amount of caching on the client is allowed */
1673 rc = ll_statfs_internal(sb, &osfs,
1674 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1679 statfs_unpack(sfs, &osfs);
1681 /* We need to downshift for all 32-bit kernels, because we can't
1682 * tell if the kernel is being called via sys_statfs64() or not.
1683 * Stop before overflowing f_bsize - in which case it is better
1684 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1685 if (sizeof(long) < 8) {
1686 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1689 osfs.os_blocks >>= 1;
1690 osfs.os_bfree >>= 1;
1691 osfs.os_bavail >>= 1;
1695 sfs->f_blocks = osfs.os_blocks;
1696 sfs->f_bfree = osfs.os_bfree;
1697 sfs->f_bavail = osfs.os_bavail;
1702 void ll_inode_size_lock(struct inode *inode)
1704 struct ll_inode_info *lli;
1706 LASSERT(!S_ISDIR(inode->i_mode));
1708 lli = ll_i2info(inode);
1709 LASSERT(lli->lli_size_sem_owner != current);
1710 down(&lli->lli_size_sem);
1711 LASSERT(lli->lli_size_sem_owner == NULL);
1712 lli->lli_size_sem_owner = current;
1715 void ll_inode_size_unlock(struct inode *inode)
1717 struct ll_inode_info *lli;
1719 lli = ll_i2info(inode);
1720 LASSERT(lli->lli_size_sem_owner == current);
1721 lli->lli_size_sem_owner = NULL;
1722 up(&lli->lli_size_sem);
1725 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1727 struct ll_inode_info *lli = ll_i2info(inode);
1728 struct mdt_body *body = md->body;
1729 struct lov_stripe_md *lsm = md->lsm;
1730 struct ll_sb_info *sbi = ll_i2sbi(inode);
1732 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1734 if (!lli->lli_has_smd &&
1735 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1736 cl_file_inode_init(inode, md);
1738 lli->lli_maxbytes = lsm->lsm_maxbytes;
1739 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1740 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1743 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1744 if (body->valid & OBD_MD_FLRMTPERM)
1745 ll_update_remote_perm(inode, md->remote_perm);
1747 #ifdef CONFIG_FS_POSIX_ACL
1748 else if (body->valid & OBD_MD_FLACL) {
1749 spin_lock(&lli->lli_lock);
1750 if (lli->lli_posix_acl)
1751 posix_acl_release(lli->lli_posix_acl);
1752 lli->lli_posix_acl = md->posix_acl;
1753 spin_unlock(&lli->lli_lock);
1756 inode->i_ino = cl_fid_build_ino(&body->fid1, ll_need_32bit_api(sbi));
1757 inode->i_generation = cl_fid_build_gen(&body->fid1);
1759 if (body->valid & OBD_MD_FLATIME) {
1760 if (body->atime > LTIME_S(inode->i_atime))
1761 LTIME_S(inode->i_atime) = body->atime;
1762 lli->lli_lvb.lvb_atime = body->atime;
1764 if (body->valid & OBD_MD_FLMTIME) {
1765 if (body->mtime > LTIME_S(inode->i_mtime)) {
1766 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1767 "to "LPU64"\n", inode->i_ino,
1768 LTIME_S(inode->i_mtime), body->mtime);
1769 LTIME_S(inode->i_mtime) = body->mtime;
1771 lli->lli_lvb.lvb_mtime = body->mtime;
1773 if (body->valid & OBD_MD_FLCTIME) {
1774 if (body->ctime > LTIME_S(inode->i_ctime))
1775 LTIME_S(inode->i_ctime) = body->ctime;
1776 lli->lli_lvb.lvb_ctime = body->ctime;
1778 if (body->valid & OBD_MD_FLMODE)
1779 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1780 if (body->valid & OBD_MD_FLTYPE)
1781 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1782 LASSERT(inode->i_mode != 0);
1783 if (S_ISREG(inode->i_mode)) {
1784 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1786 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1788 if (body->valid & OBD_MD_FLUID)
1789 inode->i_uid = body->uid;
1790 if (body->valid & OBD_MD_FLGID)
1791 inode->i_gid = body->gid;
1792 if (body->valid & OBD_MD_FLFLAGS)
1793 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1794 if (body->valid & OBD_MD_FLNLINK)
1795 set_nlink(inode, body->nlink);
1796 if (body->valid & OBD_MD_FLRDEV)
1797 inode->i_rdev = old_decode_dev(body->rdev);
1799 if (body->valid & OBD_MD_FLID) {
1800 /* FID shouldn't be changed! */
1801 if (fid_is_sane(&lli->lli_fid)) {
1802 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1803 "Trying to change FID "DFID
1804 " to the "DFID", inode %lu/%u(%p)\n",
1805 PFID(&lli->lli_fid), PFID(&body->fid1),
1806 inode->i_ino, inode->i_generation, inode);
1808 lli->lli_fid = body->fid1;
1811 LASSERT(fid_seq(&lli->lli_fid) != 0);
1813 if (body->valid & OBD_MD_FLSIZE) {
1814 if (exp_connect_som(ll_i2mdexp(inode)) &&
1815 S_ISREG(inode->i_mode)) {
1816 struct lustre_handle lockh;
1819 /* As it is possible a blocking ast has been processed
1820 * by this time, we need to check there is an UPDATE
1821 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1823 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1824 &lockh, LDLM_FL_CBPENDING);
1826 if (lli->lli_flags & (LLIF_DONE_WRITING |
1827 LLIF_EPOCH_PENDING |
1829 CERROR("ino %lu flags %u still has "
1830 "size authority! do not trust "
1831 "the size got from MDS\n",
1832 inode->i_ino, lli->lli_flags);
1834 /* Use old size assignment to avoid
1835 * deadlock bz14138 & bz14326 */
1836 i_size_write(inode, body->size);
1837 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1839 ldlm_lock_decref(&lockh, mode);
1842 /* Use old size assignment to avoid
1843 * deadlock bz14138 & bz14326 */
1844 i_size_write(inode, body->size);
1846 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1847 inode->i_ino, (unsigned long long)body->size);
1850 if (body->valid & OBD_MD_FLBLOCKS)
1851 inode->i_blocks = body->blocks;
1854 if (body->valid & OBD_MD_FLMDSCAPA) {
1855 LASSERT(md->mds_capa);
1856 ll_add_capa(inode, md->mds_capa);
1858 if (body->valid & OBD_MD_FLOSSCAPA) {
1859 LASSERT(md->oss_capa);
1860 ll_add_capa(inode, md->oss_capa);
1864 void ll_read_inode2(struct inode *inode, void *opaque)
1866 struct lustre_md *md = opaque;
1867 struct ll_inode_info *lli = ll_i2info(inode);
1870 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1871 PFID(&lli->lli_fid), inode);
1873 LASSERT(!lli->lli_has_smd);
1875 /* Core attributes from the MDS first. This is a new inode, and
1876 * the VFS doesn't zero times in the core inode so we have to do
1877 * it ourselves. They will be overwritten by either MDS or OST
1878 * attributes - we just need to make sure they aren't newer. */
1879 LTIME_S(inode->i_mtime) = 0;
1880 LTIME_S(inode->i_atime) = 0;
1881 LTIME_S(inode->i_ctime) = 0;
1883 ll_update_inode(inode, md);
1885 /* OIDEBUG(inode); */
1887 /* initializing backing dev info. */
1888 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1891 if (S_ISREG(inode->i_mode)) {
1892 struct ll_sb_info *sbi = ll_i2sbi(inode);
1893 inode->i_op = &ll_file_inode_operations;
1894 inode->i_fop = sbi->ll_fop;
1895 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1897 } else if (S_ISDIR(inode->i_mode)) {
1898 inode->i_op = &ll_dir_inode_operations;
1899 inode->i_fop = &ll_dir_operations;
1901 } else if (S_ISLNK(inode->i_mode)) {
1902 inode->i_op = &ll_fast_symlink_inode_operations;
1905 inode->i_op = &ll_special_inode_operations;
1907 init_special_inode(inode, inode->i_mode,
1914 void ll_delete_inode(struct inode *inode)
1916 struct cl_inode_info *lli = cl_i2info(inode);
1919 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1920 /* discard all dirty pages before truncating them, required by
1921 * osc_extent implementation at LU-1030. */
1922 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1924 truncate_inode_pages(&inode->i_data, 0);
1926 /* Workaround for LU-118 */
1927 if (inode->i_data.nrpages) {
1928 TREE_READ_LOCK_IRQ(&inode->i_data);
1929 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1930 LASSERTF(inode->i_data.nrpages == 0,
1931 "inode=%lu/%u(%p) nrpages=%lu, see "
1932 "http://jira.whamcloud.com/browse/LU-118\n",
1933 inode->i_ino, inode->i_generation, inode,
1934 inode->i_data.nrpages);
1936 /* Workaround end */
1938 #ifdef HAVE_SBOPS_EVICT_INODE
1939 ll_clear_inode(inode);
1946 int ll_iocontrol(struct inode *inode, struct file *file,
1947 unsigned int cmd, unsigned long arg)
1949 struct ll_sb_info *sbi = ll_i2sbi(inode);
1950 struct ptlrpc_request *req = NULL;
1955 case FSFILT_IOC_GETFLAGS: {
1956 struct mdt_body *body;
1957 struct md_op_data *op_data;
1959 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1960 0, 0, LUSTRE_OPC_ANY,
1962 if (IS_ERR(op_data))
1963 RETURN(PTR_ERR(op_data));
1965 op_data->op_valid = OBD_MD_FLFLAGS;
1966 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1967 ll_finish_md_op_data(op_data);
1969 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1973 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1975 flags = body->flags;
1977 ptlrpc_req_finished(req);
1979 RETURN(put_user(flags, (int *)arg));
1981 case FSFILT_IOC_SETFLAGS: {
1982 struct lov_stripe_md *lsm;
1983 struct obd_info oinfo = { { { 0 } } };
1984 struct md_op_data *op_data;
1986 if (get_user(flags, (int *)arg))
1989 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1990 LUSTRE_OPC_ANY, NULL);
1991 if (IS_ERR(op_data))
1992 RETURN(PTR_ERR(op_data));
1994 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1995 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1996 rc = md_setattr(sbi->ll_md_exp, op_data,
1997 NULL, 0, NULL, 0, &req, NULL);
1998 ll_finish_md_op_data(op_data);
1999 ptlrpc_req_finished(req);
2003 inode->i_flags = ll_ext_to_inode_flags(flags);
2005 lsm = ccc_inode_lsm_get(inode);
2009 OBDO_ALLOC(oinfo.oi_oa);
2011 ccc_inode_lsm_put(inode, lsm);
2015 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2016 oinfo.oi_oa->o_flags = flags;
2017 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2019 oinfo.oi_capa = ll_mdscapa_get(inode);
2020 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2021 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2022 capa_put(oinfo.oi_capa);
2023 OBDO_FREE(oinfo.oi_oa);
2024 ccc_inode_lsm_put(inode, lsm);
2026 if (rc && rc != -EPERM && rc != -EACCES)
2027 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2038 int ll_flush_ctx(struct inode *inode)
2040 struct ll_sb_info *sbi = ll_i2sbi(inode);
2042 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2044 obd_set_info_async(NULL, sbi->ll_md_exp,
2045 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2047 obd_set_info_async(NULL, sbi->ll_dt_exp,
2048 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2053 /* umount -f client means force down, don't save state */
2054 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2055 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2057 struct super_block *sb = vfsmnt->mnt_sb;
2059 void ll_umount_begin(struct super_block *sb)
2062 struct ll_sb_info *sbi = ll_s2sbi(sb);
2063 struct obd_device *obd;
2064 struct obd_ioctl_data *ioc_data;
2067 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2068 if (!(flags & MNT_FORCE)) {
2074 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2075 sb->s_count, atomic_read(&sb->s_active));
2077 obd = class_exp2obd(sbi->ll_md_exp);
2079 CERROR("Invalid MDC connection handle "LPX64"\n",
2080 sbi->ll_md_exp->exp_handle.h_cookie);
2086 obd = class_exp2obd(sbi->ll_dt_exp);
2088 CERROR("Invalid LOV connection handle "LPX64"\n",
2089 sbi->ll_dt_exp->exp_handle.h_cookie);
2095 OBD_ALLOC_PTR(ioc_data);
2097 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2098 sizeof *ioc_data, ioc_data, NULL);
2100 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2101 sizeof *ioc_data, ioc_data, NULL);
2103 OBD_FREE_PTR(ioc_data);
2107 /* Really, we'd like to wait until there are no requests outstanding,
2108 * and then continue. For now, we just invalidate the requests,
2109 * schedule() and sleep one second if needed, and hope.
2112 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2113 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2114 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2115 cfs_time_seconds(1));
2116 if (atomic_read(&vfsmnt->mnt_count) > 2)
2117 LCONSOLE_WARN("Mount still busy with %d refs! You "
2118 "may try to umount it a bit later\n",
2119 atomic_read(&vfsmnt->mnt_count));
2126 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2128 struct ll_sb_info *sbi = ll_s2sbi(sb);
2129 char *profilenm = get_profile_name(sb);
2133 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2134 read_only = *flags & MS_RDONLY;
2135 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2136 sizeof(KEY_READ_ONLY),
2137 KEY_READ_ONLY, sizeof(read_only),
2140 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2141 profilenm, read_only ?
2142 "read-only" : "read-write", err);
2147 sb->s_flags |= MS_RDONLY;
2149 sb->s_flags &= ~MS_RDONLY;
2151 if (sbi->ll_flags & LL_SBI_VERBOSE)
2152 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2153 read_only ? "read-only" : "read-write");
2158 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2159 struct super_block *sb, struct lookup_intent *it)
2161 struct ll_sb_info *sbi = NULL;
2162 struct lustre_md md;
2166 LASSERT(*inode || sb);
2167 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2168 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2169 sbi->ll_md_exp, &md);
2174 ll_update_inode(*inode, &md);
2176 LASSERT(sb != NULL);
2179 * At this point server returns to client's same fid as client
2180 * generated for creating. So using ->fid1 is okay here.
2182 LASSERT(fid_is_sane(&md.body->fid1));
2184 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2185 ll_need_32bit_api(sbi)),
2187 if (*inode == NULL || IS_ERR(*inode)) {
2188 #ifdef CONFIG_FS_POSIX_ACL
2190 posix_acl_release(md.posix_acl);
2191 md.posix_acl = NULL;
2194 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2196 CERROR("new_inode -fatal: rc %d\n", rc);
2201 /* Handling piggyback layout lock.
2202 * Layout lock can be piggybacked by getattr and open request.
2203 * The lsm can be applied to inode only if it comes with a layout lock
2204 * otherwise correct layout may be overwritten, for example:
2205 * 1. proc1: mdt returns a lsm but not granting layout
2206 * 2. layout was changed by another client
2207 * 3. proc2: refresh layout and layout lock granted
2208 * 4. proc1: to apply a stale layout */
2209 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2210 struct lustre_handle lockh;
2211 struct ldlm_lock *lock;
2213 lockh.cookie = it->d.lustre.it_lock_handle;
2214 lock = ldlm_handle2lock(&lockh);
2215 LASSERT(lock != NULL);
2216 if (ldlm_has_layout(lock)) {
2217 struct cl_object_conf conf;
2219 memset(&conf, 0, sizeof(conf));
2220 conf.coc_opc = OBJECT_CONF_SET;
2221 conf.coc_inode = *inode;
2222 conf.coc_lock = lock;
2223 conf.u.coc_md = &md;
2224 (void)ll_layout_conf(*inode, &conf);
2226 LDLM_LOCK_PUT(lock);
2231 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2232 md_free_lustre_md(sbi->ll_md_exp, &md);
2236 int ll_obd_statfs(struct inode *inode, void *arg)
2238 struct ll_sb_info *sbi = NULL;
2239 struct obd_export *exp;
2241 struct obd_ioctl_data *data = NULL;
2246 if (!inode || !(sbi = ll_i2sbi(inode)))
2247 GOTO(out_statfs, rc = -EINVAL);
2249 rc = obd_ioctl_getdata(&buf, &len, arg);
2251 GOTO(out_statfs, rc);
2254 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2255 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2256 GOTO(out_statfs, rc = -EINVAL);
2258 if (data->ioc_inllen1 != sizeof(__u32) ||
2259 data->ioc_inllen2 != sizeof(__u32) ||
2260 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2261 data->ioc_plen2 != sizeof(struct obd_uuid))
2262 GOTO(out_statfs, rc = -EINVAL);
2264 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2265 if (type & LL_STATFS_LMV)
2266 exp = sbi->ll_md_exp;
2267 else if (type & LL_STATFS_LOV)
2268 exp = sbi->ll_dt_exp;
2270 GOTO(out_statfs, rc = -ENODEV);
2272 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2273 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2275 GOTO(out_statfs, rc);
2278 obd_ioctl_freedata(buf, len);
2282 int ll_process_config(struct lustre_cfg *lcfg)
2286 struct lprocfs_static_vars lvars;
2290 lprocfs_llite_init_vars(&lvars);
2292 /* The instance name contains the sb: lustre-client-aacfe000 */
2293 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2294 if (!ptr || !*(++ptr))
2296 if (sscanf(ptr, "%lx", &x) != 1)
2299 /* This better be a real Lustre superblock! */
2300 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2302 /* Note we have not called client_common_fill_super yet, so
2303 proc fns must be able to handle that! */
2304 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2311 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2312 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2313 struct inode *i1, struct inode *i2,
2314 const char *name, int namelen,
2315 int mode, __u32 opc, void *data)
2317 LASSERT(i1 != NULL);
2319 if (namelen > ll_i2sbi(i1)->ll_namelen)
2320 return ERR_PTR(-ENAMETOOLONG);
2322 if (op_data == NULL)
2323 OBD_ALLOC_PTR(op_data);
2325 if (op_data == NULL)
2326 return ERR_PTR(-ENOMEM);
2328 ll_i2gids(op_data->op_suppgids, i1, i2);
2329 op_data->op_fid1 = *ll_inode2fid(i1);
2330 op_data->op_capa1 = ll_mdscapa_get(i1);
2333 op_data->op_fid2 = *ll_inode2fid(i2);
2334 op_data->op_capa2 = ll_mdscapa_get(i2);
2336 fid_zero(&op_data->op_fid2);
2337 op_data->op_capa2 = NULL;
2340 op_data->op_name = name;
2341 op_data->op_namelen = namelen;
2342 op_data->op_mode = mode;
2343 op_data->op_mod_time = cfs_time_current_sec();
2344 op_data->op_fsuid = cfs_curproc_fsuid();
2345 op_data->op_fsgid = cfs_curproc_fsgid();
2346 op_data->op_cap = cfs_curproc_cap_pack();
2347 op_data->op_bias = 0;
2348 op_data->op_cli_flags = 0;
2349 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2350 filename_is_volatile(name, namelen, NULL))
2351 op_data->op_bias |= MDS_CREATE_VOLATILE;
2352 op_data->op_opc = opc;
2353 op_data->op_mds = 0;
2354 op_data->op_data = data;
2356 /* If the file is being opened after mknod() (normally due to NFS)
2357 * try to use the default stripe data from parent directory for
2358 * allocating OST objects. Try to pass the parent FID to MDS. */
2359 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2360 !ll_i2info(i2)->lli_has_smd) {
2361 struct ll_inode_info *lli = ll_i2info(i2);
2363 spin_lock(&lli->lli_lock);
2364 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2365 op_data->op_fid1 = lli->lli_pfid;
2366 spin_unlock(&lli->lli_lock);
2367 /** We ignore parent's capability temporary. */
2370 /* When called by ll_setattr_raw, file is i1. */
2371 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2372 op_data->op_bias |= MDS_DATA_MODIFIED;
2377 void ll_finish_md_op_data(struct md_op_data *op_data)
2379 capa_put(op_data->op_capa1);
2380 capa_put(op_data->op_capa2);
2381 OBD_FREE_PTR(op_data);
2384 #ifdef HAVE_SUPEROPS_USE_DENTRY
2385 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2387 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2390 struct ll_sb_info *sbi;
2392 #ifdef HAVE_SUPEROPS_USE_DENTRY
2393 LASSERT((seq != NULL) && (dentry != NULL));
2394 sbi = ll_s2sbi(dentry->d_sb);
2396 LASSERT((seq != NULL) && (vfs != NULL));
2397 sbi = ll_s2sbi(vfs->mnt_sb);
2400 if (sbi->ll_flags & LL_SBI_NOLCK)
2401 seq_puts(seq, ",nolock");
2403 if (sbi->ll_flags & LL_SBI_FLOCK)
2404 seq_puts(seq, ",flock");
2406 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2407 seq_puts(seq, ",localflock");
2409 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2410 seq_puts(seq, ",user_xattr");
2412 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2413 seq_puts(seq, ",lazystatfs");
2415 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2416 seq_puts(seq, ",user_fid2path");
2422 * Get obd name by cmd, and copy out to user space
2424 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2426 struct ll_sb_info *sbi = ll_i2sbi(inode);
2427 struct obd_device *obd;
2430 if (cmd == OBD_IOC_GETDTNAME)
2431 obd = class_exp2obd(sbi->ll_dt_exp);
2432 else if (cmd == OBD_IOC_GETMDNAME)
2433 obd = class_exp2obd(sbi->ll_md_exp);
2440 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2441 strlen(obd->obd_name) + 1))
2448 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2449 * fsname will be returned in this buffer; otherwise, a static buffer will be
2450 * used to store the fsname and returned to caller.
2452 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2454 static char fsname_static[MTI_NAME_MAXLEN];
2455 struct lustre_sb_info *lsi = s2lsi(sb);
2460 /* this means the caller wants to use static buffer
2461 * and it doesn't care about race. Usually this is
2462 * in error reporting path */
2463 buf = fsname_static;
2464 buflen = sizeof(fsname_static);
2467 len = strlen(lsi->lsi_lmd->lmd_profile);
2468 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2469 if (ptr && (strcmp(ptr, "-client") == 0))
2472 if (unlikely(len >= buflen))
2474 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2480 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2484 #ifdef HAVE_FS_STRUCT_USE_PATH
2488 p.mnt = current->fs->root.mnt;
2490 path = d_path(&p, buf, bufsize);
2493 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2499 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2501 char *buf, *path = NULL;
2502 struct dentry *dentry = NULL;
2503 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2505 /* this can be called inside spin lock so use GFP_ATOMIC. */
2506 buf = (char *)__get_free_page(GFP_ATOMIC);
2508 dentry = d_find_alias(page->mapping->host);
2510 path = ll_d_path(dentry, buf, PAGE_SIZE);
2513 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2514 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2515 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2516 PFID(&obj->cob_header.coh_lu.loh_fid),
2517 (path && !IS_ERR(path)) ? path : "", ioret);
2523 free_page((unsigned long)buf);