1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <obd_cksum.h>
57 #include <lustre_cache.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 LIST_HEAD(ll_super_blocks);
63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) ffz(~(n))
72 static inline void ll_pglist_fini(struct ll_sb_info *sbi)
77 if (sbi->ll_pglist == NULL)
80 for_each_possible_cpu(i) {
81 page = sbi->ll_pglist[i]->llpd_page;
83 sbi->ll_pglist[i] = NULL;
88 OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
89 sbi->ll_pglist = NULL;
92 static inline int ll_pglist_init(struct ll_sb_info *sbi)
94 struct ll_pglist_data *pd;
99 OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
100 if (sbi->ll_pglist == NULL)
103 budget = sbi->ll_async_page_max / num_online_cpus();
104 for_each_possible_cpu(i) {
105 struct page *page = alloc_pages_node(cpu_to_node(i),
112 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
115 pd = (struct ll_pglist_data *)(page_address(page) + color);
116 memset(pd, 0, sizeof(*pd));
117 spin_lock_init(&pd->llpd_lock);
118 INIT_LIST_HEAD(&pd->llpd_list);
120 pd->llpd_budget = budget;
122 pd->llpd_page = page;
123 atomic_set(&pd->llpd_sample_count, 0);
124 sbi->ll_pglist[i] = pd;
125 color += L1_CACHE_ALIGN(sizeof(*pd));
131 static struct ll_sb_info *ll_init_sbi(void)
133 struct ll_sb_info *sbi = NULL;
140 OBD_ALLOC(sbi, sizeof(*sbi));
144 OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
145 if (sbi->ll_async_page_sample == NULL)
148 spin_lock_init(&sbi->ll_lock);
149 spin_lock_init(&sbi->ll_lco.lco_lock);
150 spin_lock_init(&sbi->ll_pp_extent_lock);
151 spin_lock_init(&sbi->ll_process_lock);
152 sbi->ll_rw_stats_on = 0;
155 pages = si.totalram - si.totalhigh;
156 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
157 #ifdef HAVE_BGL_SUPPORT
158 sbi->ll_async_page_max = pages / 4;
160 sbi->ll_async_page_max = pages / 2;
163 sbi->ll_async_page_max = (pages / 4) * 3;
165 lcounter_init(&sbi->ll_async_page_count);
166 spin_lock_init(&sbi->ll_async_page_reblnc_lock);
167 sbi->ll_async_page_sample_max = 64 * num_online_cpus();
168 sbi->ll_async_page_reblnc_count = 0;
169 sbi->ll_async_page_clock_hand = 0;
170 if (ll_pglist_init(sbi))
173 sbi->ll_ra_info.ra_max_pages = min(pages / 32,
174 SBI_DEFAULT_READAHEAD_MAX);
175 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
176 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
177 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
178 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
179 INIT_LIST_HEAD(&sbi->ll_conn_chain);
180 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
182 ll_generate_random_uuid(uuid);
183 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
184 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
186 spin_lock(&ll_sb_lock);
187 list_add_tail(&sbi->ll_list, &ll_super_blocks);
188 spin_unlock(&ll_sb_lock);
190 #ifdef ENABLE_LLITE_CHECKSUM
191 sbi->ll_flags |= LL_SBI_CHECKSUM;
194 #ifdef HAVE_LRU_RESIZE_SUPPORT
195 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
198 #ifdef HAVE_EXPORT___IGET
199 INIT_LIST_HEAD(&sbi->ll_deathrow);
200 spin_lock_init(&sbi->ll_deathrow_lock);
202 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
203 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
204 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
207 /* metadata statahead is enabled by default */
208 sbi->ll_sa_max = LL_SA_RPC_DEF;
213 if (sbi->ll_async_page_sample)
214 OBD_FREE(sbi->ll_async_page_sample,
215 sizeof(long) * num_possible_cpus());
217 OBD_FREE(sbi, sizeof(*sbi));
221 void ll_free_sbi(struct super_block *sb)
223 struct ll_sb_info *sbi = ll_s2sbi(sb);
228 spin_lock(&ll_sb_lock);
229 list_del(&sbi->ll_list);
230 spin_unlock(&ll_sb_lock);
231 lcounter_destroy(&sbi->ll_async_page_count);
232 OBD_FREE(sbi->ll_async_page_sample,
233 sizeof(long) * num_possible_cpus());
234 OBD_FREE(sbi, sizeof(*sbi));
239 static struct dentry_operations ll_d_root_ops = {
240 #ifdef DCACHE_LUSTRE_INVALID
241 .d_compare = ll_dcompare,
245 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
246 * us to make MDS RPCs with large enough reply buffers to hold the
247 * maximum-sized (= maximum striped) EA and cookie without having to
248 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
249 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
251 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
252 __u32 valsize = sizeof(struct lov_desc);
253 int rc, easize, def_easize, cookiesize;
254 struct lov_desc desc;
258 rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
259 &valsize, &desc, NULL);
263 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
264 lsm.lsm_stripe_count = stripes;
265 easize = obd_size_diskmd(dt_exp, &lsm);
267 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
268 def_easize = obd_size_diskmd(dt_exp, &lsm);
270 cookiesize = stripes * sizeof(struct llog_cookie);
272 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
275 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
279 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
281 struct inode *root = 0;
282 struct ll_sb_info *sbi = ll_s2sbi(sb);
283 struct obd_device *obd;
284 struct lu_fid rootfid;
285 struct obd_capa *oc = NULL;
286 struct obd_statfs osfs;
287 struct ptlrpc_request *request = NULL;
288 struct lustre_handle dt_conn = {0, };
289 struct lustre_handle md_conn = {0, };
290 struct obd_connect_data *data = NULL;
291 struct obd_uuid *uuid;
292 struct lustre_md lmd;
294 int size, err, checksum;
297 obd = class_name2obd(md);
299 CERROR("MD %s: not setup or attached\n", md);
307 if (proc_lustre_fs_root) {
308 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
311 CERROR("could not register mount in /proc/fs/lustre\n");
314 /* indicate the features supported by this client */
315 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
316 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
317 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
318 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
319 OBD_CONNECT_FID | OBD_CONNECT_AT;
321 #ifdef HAVE_LRU_RESIZE_SUPPORT
322 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
323 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
325 #ifdef CONFIG_FS_POSIX_ACL
326 data->ocd_connect_flags |= OBD_CONNECT_ACL;
328 data->ocd_ibits_known = MDS_INODELOCK_FULL;
329 data->ocd_version = LUSTRE_VERSION_CODE;
331 if (sb->s_flags & MS_RDONLY)
332 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
333 if (sbi->ll_flags & LL_SBI_USER_XATTR)
334 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
336 #ifdef HAVE_MS_FLOCK_LOCK
337 /* force vfs to use lustre handler for flock() calls - bug 10743 */
338 sb->s_flags |= MS_FLOCK_LOCK;
341 if (sbi->ll_flags & LL_SBI_FLOCK)
342 sbi->ll_fop = &ll_file_operations_flock;
343 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
344 sbi->ll_fop = &ll_file_operations;
346 sbi->ll_fop = &ll_file_operations_noflock;
349 data->ocd_connect_flags |= OBD_CONNECT_REAL;
350 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
351 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
352 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
354 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
355 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
358 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
360 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
361 "recovery, of which this client is not a "
362 "part. Please wait for recovery to complete,"
363 " abort, or time out.\n", md);
366 CERROR("cannot connect to %s: rc = %d\n", md, err);
369 sbi->ll_md_exp = class_conn2export(&md_conn);
371 err = obd_fid_init(sbi->ll_md_exp);
373 CERROR("Can't init metadata layer FID infrastructure, "
378 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
380 GOTO(out_md_fid, err);
382 size = sizeof(*data);
383 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
384 KEY_CONN_DATA, &size, data, NULL);
386 CERROR("Get connect data failed: %d \n", err);
390 LASSERT(osfs.os_bsize);
391 sb->s_blocksize = osfs.os_bsize;
392 sb->s_blocksize_bits = log2(osfs.os_bsize);
393 sb->s_magic = LL_SUPER_MAGIC;
395 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
396 * retval = in_file->f_op->sendfile(...);
398 * retval = -EOVERFLOW;
400 * it will check if *ppos is greater than max. However, max equals to
401 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
402 * has been defined as a signed long long ineger in linux kernel. */
403 #if BITS_PER_LONG == 64
404 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
406 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
408 sbi->ll_namelen = osfs.os_namelen;
409 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
411 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
412 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
413 LCONSOLE_INFO("Disabling user_xattr feature because "
414 "it is not supported on the server\n");
415 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
418 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
420 sb->s_flags |= MS_POSIXACL;
422 sbi->ll_flags |= LL_SBI_ACL;
424 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
426 sb->s_flags &= ~MS_POSIXACL;
428 sbi->ll_flags &= ~LL_SBI_ACL;
431 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
432 sbi->ll_flags |= LL_SBI_JOIN;
434 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
435 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
436 /* sometimes local client claims to be remote, but mdt
437 * will disagree when client gss not applied. */
438 LCONSOLE_INFO("client claims to be remote, but server "
439 "rejected, forced to be local.\n");
440 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
443 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
444 /* with gss applied, remote client can not claim to be
445 * local, so mdt maybe force client to be remote. */
446 LCONSOLE_INFO("client claims to be local, but server "
447 "rejected, forced to be remote.\n");
448 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
452 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
453 LCONSOLE_INFO("client enabled MDS capability!\n");
454 sbi->ll_flags |= LL_SBI_MDS_CAPA;
457 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
458 LCONSOLE_INFO("client enabled OSS capability!\n");
459 sbi->ll_flags |= LL_SBI_OSS_CAPA;
462 obd = class_name2obd(dt);
464 CERROR("DT %s: not setup or attached\n", dt);
465 GOTO(out_md_fid, err = -ENODEV);
468 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
469 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
470 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
471 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
473 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
474 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
476 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
477 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
478 * disabled by default, because it can still be enabled on the
479 * fly via /proc. As a consequence, we still need to come to an
480 * agreement on the supported algorithms at connect time */
481 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
483 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
484 data->ocd_cksum_types = OBD_CKSUM_ADLER;
486 /* send the list of supported checksum types */
487 data->ocd_cksum_types = OBD_CKSUM_ALL;
490 #ifdef HAVE_LRU_RESIZE_SUPPORT
491 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
493 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
494 "ocd_grant: %d\n", data->ocd_connect_flags,
495 data->ocd_version, data->ocd_grant);
497 obd->obd_upcall.onu_owner = &sbi->ll_lco;
498 obd->obd_upcall.onu_upcall = ll_ocd_update;
499 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
501 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
503 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
504 "recovery, of which this client is not a "
505 "part. Please wait for recovery to "
506 "complete, abort, or time out.\n", dt);
507 GOTO(out_md_fid, err);
509 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
510 GOTO(out_md_fid, err);
513 sbi->ll_dt_exp = class_conn2export(&dt_conn);
515 err = obd_fid_init(sbi->ll_dt_exp);
517 CERROR("Can't init data layer FID infrastructure, "
522 spin_lock(&sbi->ll_lco.lco_lock);
523 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
524 spin_unlock(&sbi->ll_lco.lco_lock);
526 err = obd_register_page_removal_cb(sbi->ll_dt_exp,
530 CERROR("cannot register page removal callback: rc = %d\n",err);
533 err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
534 ll_extent_lock_cancel_cb);
536 CERROR("cannot register lock cancel callback: rc = %d\n", err);
537 GOTO(out_page_rm_cb, err);
540 err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
542 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
543 GOTO(out_lock_cn_cb, err);
546 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
547 0, NULL, NULL, NULL, 0, NULL);
549 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
550 "filesystem. There must be at least one "
551 "active OST for a client to start.\n");
552 GOTO(out_lock_cn_cb, err);
555 if (!ll_async_page_slab) {
556 ll_async_page_slab_size =
557 size_round(sizeof(struct ll_async_page)) + err;
558 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
559 ll_async_page_slab_size,
561 if (!ll_async_page_slab)
562 GOTO(out_lock_cn_cb, err = -ENOMEM);
565 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
567 CERROR("cannot mds_connect: rc = %d\n", err);
568 GOTO(out_lock_cn_cb, err);
570 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
571 sbi->ll_root_fid = rootfid;
573 sb->s_op = &lustre_super_operations;
574 sb->s_export_op = &lustre_export_operations;
577 * XXX: move this to after cbd setup? */
578 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
579 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
580 valid |= OBD_MD_FLRMTPERM;
581 else if (sbi->ll_flags & LL_SBI_ACL)
582 valid |= OBD_MD_FLACL;
584 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
588 CERROR("md_getattr failed for root: rc = %d\n", err);
589 GOTO(out_lock_cn_cb, err);
591 memset(&lmd, 0, sizeof(lmd));
592 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
593 sbi->ll_md_exp, &lmd);
595 CERROR("failed to understand root inode md: rc = %d\n", err);
596 ptlrpc_req_finished (request);
597 GOTO(out_lock_cn_cb, err);
600 LASSERT(fid_is_sane(&sbi->ll_root_fid));
601 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
602 md_free_lustre_md(sbi->ll_md_exp, &lmd);
603 ptlrpc_req_finished(request);
605 if (root == NULL || is_bad_inode(root)) {
607 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
608 #ifdef CONFIG_FS_POSIX_ACL
610 posix_acl_release(lmd.posix_acl);
611 lmd.posix_acl = NULL;
614 CERROR("lustre_lite: bad iget4 for root\n");
615 GOTO(out_root, err = -EBADF);
618 err = ll_close_thread_start(&sbi->ll_lcq);
620 CERROR("cannot start close thread: rc %d\n", err);
624 #ifdef CONFIG_FS_POSIX_ACL
625 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
626 rct_init(&sbi->ll_rct);
627 et_init(&sbi->ll_et);
631 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
632 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
633 KEY_CHECKSUM, sizeof(checksum), &checksum,
636 sb->s_root = d_alloc_root(root);
638 OBD_FREE(data, sizeof(*data));
640 sb->s_root->d_op = &ll_d_root_ops;
642 sbi->ll_sdev_orig = sb->s_dev;
644 /* We set sb->s_dev equal on all lustre clients in order to support
645 * NFS export clustering. NFSD requires that the FSID be the same
647 /* s_dev is also used in lt_compare() to compare two fs, but that is
648 * only a node-local comparison. */
649 uuid = obd_get_uuid(sbi->ll_md_exp);
651 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
658 obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
659 ll_extent_lock_cancel_cb);
661 obd_unregister_page_removal_cb(sbi->ll_dt_exp,
663 obd_fid_fini(sbi->ll_dt_exp);
665 obd_disconnect(sbi->ll_dt_exp);
666 sbi->ll_dt_exp = NULL;
668 obd_fid_fini(sbi->ll_md_exp);
670 obd_disconnect(sbi->ll_md_exp);
671 sbi->ll_md_exp = NULL;
675 lprocfs_unregister_mountpoint(sbi);
679 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
683 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
685 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
686 KEY_MAX_EASIZE, &size, lmmsize, NULL);
688 CERROR("Get max mdsize error rc %d \n", rc);
693 void ll_dump_inode(struct inode *inode)
695 struct list_head *tmp;
696 int dentry_count = 0;
698 LASSERT(inode != NULL);
700 list_for_each(tmp, &inode->i_dentry)
703 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
704 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
705 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
708 void lustre_dump_dentry(struct dentry *dentry, int recur)
710 struct list_head *tmp;
713 LASSERT(dentry != NULL);
715 list_for_each(tmp, &dentry->d_subdirs)
718 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
719 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
720 dentry->d_name.len, dentry->d_name.name,
721 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
722 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
723 dentry->d_flags, dentry->d_fsdata, subdirs);
724 if (dentry->d_inode != NULL)
725 ll_dump_inode(dentry->d_inode);
730 list_for_each(tmp, &dentry->d_subdirs) {
731 struct dentry *d = list_entry(tmp, struct dentry, d_child);
732 lustre_dump_dentry(d, recur - 1);
736 #ifdef HAVE_EXPORT___IGET
737 static void prune_dir_dentries(struct inode *inode)
739 struct dentry *dentry, *prev = NULL;
741 /* due to lustre specific logic, a directory
742 * can have few dentries - a bug from VFS POV */
744 spin_lock(&dcache_lock);
745 if (!list_empty(&inode->i_dentry)) {
746 dentry = list_entry(inode->i_dentry.prev,
747 struct dentry, d_alias);
748 /* in order to prevent infinite loops we
749 * break if previous dentry is busy */
750 if (dentry != prev) {
753 spin_unlock(&dcache_lock);
755 /* try to kill all child dentries */
757 shrink_dcache_parent(dentry);
758 unlock_dentry(dentry);
761 /* now try to get rid of current dentry */
762 d_prune_aliases(inode);
766 spin_unlock(&dcache_lock);
769 static void prune_deathrow_one(struct ll_inode_info *lli)
771 struct inode *inode = ll_info2i(lli);
773 /* first, try to drop any dentries - they hold a ref on the inode */
774 if (S_ISDIR(inode->i_mode))
775 prune_dir_dentries(inode);
777 d_prune_aliases(inode);
780 /* if somebody still uses it, leave it */
781 LASSERT(atomic_read(&inode->i_count) > 0);
782 if (atomic_read(&inode->i_count) > 1)
785 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
786 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
788 /* seems nobody uses it anymore */
796 static void prune_deathrow(struct ll_sb_info *sbi, int try)
798 struct ll_inode_info *lli;
802 if (need_resched() && try)
806 if (!spin_trylock(&sbi->ll_deathrow_lock))
809 spin_lock(&sbi->ll_deathrow_lock);
814 if (!list_empty(&sbi->ll_deathrow)) {
815 lli = list_entry(sbi->ll_deathrow.next,
816 struct ll_inode_info,
818 list_del_init(&lli->lli_dead_list);
819 if (!list_empty(&sbi->ll_deathrow))
822 spin_unlock(&sbi->ll_deathrow_lock);
825 prune_deathrow_one(lli);
827 } while (empty == 0);
829 #else /* !HAVE_EXPORT___IGET */
830 #define prune_deathrow(sbi, try) do {} while (0)
831 #endif /* HAVE_EXPORT___IGET */
833 void client_common_put_super(struct super_block *sb)
835 struct ll_sb_info *sbi = ll_s2sbi(sb);
838 #ifdef CONFIG_FS_POSIX_ACL
839 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
840 et_fini(&sbi->ll_et);
841 rct_fini(&sbi->ll_rct);
845 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
847 ll_close_thread_shutdown(sbi->ll_lcq);
849 /* destroy inodes in deathrow */
850 prune_deathrow(sbi, 0);
852 list_del(&sbi->ll_conn_chain);
854 obd_fid_fini(sbi->ll_dt_exp);
855 obd_disconnect(sbi->ll_dt_exp);
856 sbi->ll_dt_exp = NULL;
858 lprocfs_unregister_mountpoint(sbi);
860 obd_fid_fini(sbi->ll_md_exp);
861 obd_disconnect(sbi->ll_md_exp);
862 sbi->ll_md_exp = NULL;
867 void ll_kill_super(struct super_block *sb)
869 struct ll_sb_info *sbi;
874 if (!(sb->s_flags & MS_ACTIVE))
878 /* we need restore s_dev from changed for clustred NFS before put_super
879 * because new kernels have cached s_dev and change sb->s_dev in
880 * put_super not affected real removing devices */
882 sb->s_dev = sbi->ll_sdev_orig;
886 char *ll_read_opt(const char *opt, char *data)
892 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
893 if (strncmp(opt, data, strlen(opt)))
895 if ((value = strchr(data, '=')) == NULL)
899 OBD_ALLOC(retval, strlen(value) + 1);
901 CERROR("out of memory!\n");
905 memcpy(retval, value, strlen(value)+1);
906 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
910 static inline int ll_set_opt(const char *opt, char *data, int fl)
912 if (strncmp(opt, data, strlen(opt)) != 0)
918 /* non-client-specific mount options are parsed in lmd_parse */
919 static int ll_options(char *options, int *flags)
922 char *s1 = options, *s2;
928 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
931 CDEBUG(D_SUPER, "next opt=%s\n", s1);
932 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
937 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
942 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
947 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
952 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
957 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
962 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
964 /* Ignore deprecated mount option. The client will
965 * always try to mount with ACL support, whether this
966 * is used depends on whether server supports it. */
969 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
973 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
979 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
984 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
989 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
994 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
1000 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1006 s2 = strchr(s1, ',');
1014 void ll_lli_init(struct ll_inode_info *lli)
1016 lli->lli_inode_magic = LLI_INODE_MAGIC;
1017 sema_init(&lli->lli_size_sem, 1);
1018 sema_init(&lli->lli_write_sem, 1);
1020 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1021 spin_lock_init(&lli->lli_lock);
1022 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1023 INIT_LIST_HEAD(&lli->lli_close_list);
1024 lli->lli_inode_magic = LLI_INODE_MAGIC;
1025 sema_init(&lli->lli_och_sem, 1);
1026 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1027 lli->lli_mds_exec_och = NULL;
1028 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1029 lli->lli_open_fd_exec_count = 0;
1030 INIT_LIST_HEAD(&lli->lli_dead_list);
1031 lli->lli_remote_perms = NULL;
1032 lli->lli_rmtperm_utime = 0;
1033 sema_init(&lli->lli_rmtperm_sem, 1);
1034 INIT_LIST_HEAD(&lli->lli_oss_capas);
1037 int ll_fill_super(struct super_block *sb)
1039 struct lustre_profile *lprof;
1040 struct lustre_sb_info *lsi = s2lsi(sb);
1041 struct ll_sb_info *sbi;
1042 char *dt = NULL, *md = NULL;
1043 char *profilenm = get_profile_name(sb);
1044 struct config_llog_instance cfg = {0, };
1045 char ll_instance[sizeof(sb) * 2 + 1];
1049 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1053 /* client additional sb info */
1054 lsi->lsi_llsbi = sbi = ll_init_sbi();
1060 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1062 GOTO(out_free, err);
1064 /* Generate a string unique to this super, in case some joker tries
1065 to mount the same fs at two mount points.
1066 Use the address of the super itself.*/
1067 sprintf(ll_instance, "%p", sb);
1068 cfg.cfg_instance = ll_instance;
1069 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1071 /* set up client obds */
1072 err = lustre_process_log(sb, profilenm, &cfg);
1074 CERROR("Unable to process log: %d\n", err);
1075 GOTO(out_free, err);
1078 lprof = class_get_profile(profilenm);
1079 if (lprof == NULL) {
1080 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1081 " read from the MGS. Does that filesystem "
1082 "exist?\n", profilenm);
1083 GOTO(out_free, err = -EINVAL);
1085 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1086 lprof->lp_md, lprof->lp_dt);
1088 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1089 strlen(ll_instance) + 2);
1091 GOTO(out_free, err = -ENOMEM);
1092 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1094 OBD_ALLOC(md, strlen(lprof->lp_md) +
1095 strlen(ll_instance) + 2);
1097 GOTO(out_free, err = -ENOMEM);
1098 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1100 /* connections, registrations, sb setup */
1101 err = client_common_fill_super(sb, md, dt);
1105 OBD_FREE(md, strlen(md) + 1);
1107 OBD_FREE(dt, strlen(dt) + 1);
1111 LCONSOLE_WARN("Client %s has started\n", profilenm);
1114 } /* ll_fill_super */
1117 void ll_put_super(struct super_block *sb)
1119 struct config_llog_instance cfg;
1120 char ll_instance[sizeof(sb) * 2 + 1];
1121 struct obd_device *obd;
1122 struct lustre_sb_info *lsi = s2lsi(sb);
1123 struct ll_sb_info *sbi = ll_s2sbi(sb);
1124 char *profilenm = get_profile_name(sb);
1125 int force = 1, next;
1128 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1130 ll_print_capa_stat(sbi);
1132 sprintf(ll_instance, "%p", sb);
1133 cfg.cfg_instance = ll_instance;
1134 lustre_end_log(sb, NULL, &cfg);
1136 if (sbi->ll_md_exp) {
1137 obd = class_exp2obd(sbi->ll_md_exp);
1139 force = obd->obd_force;
1142 /* We need to set force before the lov_disconnect in
1143 lustre_common_put_super, since l_d cleans up osc's as well. */
1146 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1148 obd->obd_force = force;
1153 /* Only if client_common_fill_super succeeded */
1154 client_common_put_super(sb);
1157 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1158 class_manual_cleanup(obd);
1162 class_del_profile(profilenm);
1165 lsi->lsi_llsbi = NULL;
1167 lustre_common_put_super(sb);
1169 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1174 } /* client_put_super */
1176 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1178 #if defined(HAVE_CACHE_RETURN_INT)
1183 ll_shrink_cache(int priority, unsigned int gfp_mask)
1185 struct ll_sb_info *sbi;
1188 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1189 count += llap_shrink_cache(sbi, priority);
1191 #if defined(HAVE_CACHE_RETURN_INT)
1196 struct cache_definition ll_cache_definition = {
1197 .name = "llap_cache",
1198 .shrink = ll_shrink_cache
1200 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1202 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1204 struct inode *inode = NULL;
1205 /* NOTE: we depend on atomic igrab() -bzzz */
1206 lock_res_and_lock(lock);
1207 if (lock->l_ast_data) {
1208 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1209 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1210 inode = igrab(lock->l_ast_data);
1212 inode = lock->l_ast_data;
1213 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1215 lock, __FILE__, __func__, __LINE__,
1216 "l_ast_data %p is bogus: magic %08x",
1217 lock->l_ast_data, lli->lli_inode_magic);
1221 unlock_res_and_lock(lock);
1225 static int null_if_equal(struct ldlm_lock *lock, void *data)
1227 if (data == lock->l_ast_data) {
1228 lock->l_ast_data = NULL;
1230 if (lock->l_req_mode != lock->l_granted_mode)
1231 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1234 return LDLM_ITER_CONTINUE;
1237 void ll_clear_inode(struct inode *inode)
1239 struct ll_inode_info *lli = ll_i2info(inode);
1240 struct ll_sb_info *sbi = ll_i2sbi(inode);
1243 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1244 inode->i_generation, inode);
1246 if (S_ISDIR(inode->i_mode)) {
1247 /* these should have been cleared in ll_file_release */
1248 LASSERT(lli->lli_sai == NULL);
1249 LASSERT(lli->lli_opendir_key == NULL);
1250 LASSERT(lli->lli_opendir_pid == 0);
1253 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1254 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1255 null_if_equal, inode);
1257 LASSERT(!lli->lli_open_fd_write_count);
1258 LASSERT(!lli->lli_open_fd_read_count);
1259 LASSERT(!lli->lli_open_fd_exec_count);
1261 if (lli->lli_mds_write_och)
1262 ll_md_real_close(inode, FMODE_WRITE);
1263 if (lli->lli_mds_exec_och)
1264 ll_md_real_close(inode, FMODE_EXEC);
1265 if (lli->lli_mds_read_och)
1266 ll_md_real_close(inode, FMODE_READ);
1269 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1270 null_if_equal, inode);
1272 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1273 lli->lli_smd = NULL;
1276 if (lli->lli_symlink_name) {
1277 OBD_FREE(lli->lli_symlink_name,
1278 strlen(lli->lli_symlink_name) + 1);
1279 lli->lli_symlink_name = NULL;
1282 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1283 LASSERT(lli->lli_posix_acl == NULL);
1284 if (lli->lli_remote_perms) {
1285 free_rmtperm_hash(lli->lli_remote_perms);
1286 lli->lli_remote_perms = NULL;
1289 #ifdef CONFIG_FS_POSIX_ACL
1290 else if (lli->lli_posix_acl) {
1291 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1292 LASSERT(lli->lli_remote_perms == NULL);
1293 posix_acl_release(lli->lli_posix_acl);
1294 lli->lli_posix_acl = NULL;
1297 lli->lli_inode_magic = LLI_INODE_DEAD;
1299 #ifdef HAVE_EXPORT___IGET
1300 spin_lock(&sbi->ll_deathrow_lock);
1301 list_del_init(&lli->lli_dead_list);
1302 spin_unlock(&sbi->ll_deathrow_lock);
1304 ll_clear_inode_capas(inode);
1309 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1310 struct md_open_data **mod)
1312 struct lustre_md md;
1313 struct ll_sb_info *sbi = ll_i2sbi(inode);
1314 struct ptlrpc_request *request = NULL;
1318 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1319 LUSTRE_OPC_ANY, NULL);
1320 if (IS_ERR(op_data))
1321 RETURN(PTR_ERR(op_data));
1323 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1326 ptlrpc_req_finished(request);
1327 if (rc == -ENOENT) {
1329 /* Unlinked special device node? Or just a race?
1330 * Pretend we done everything. */
1331 if (!S_ISREG(inode->i_mode) &&
1332 !S_ISDIR(inode->i_mode))
1333 rc = inode_setattr(inode, &op_data->op_attr);
1334 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1335 CERROR("md_setattr fails: rc = %d\n", rc);
1340 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1341 sbi->ll_md_exp, &md);
1343 ptlrpc_req_finished(request);
1347 /* We call inode_setattr to adjust timestamps.
1348 * If there is at least some data in file, we cleared ATTR_SIZE
1349 * above to avoid invoking vmtruncate, otherwise it is important
1350 * to call vmtruncate in inode_setattr to update inode->i_size
1352 rc = inode_setattr(inode, &op_data->op_attr);
1354 /* Extract epoch data if obtained. */
1355 op_data->op_handle = md.body->handle;
1356 op_data->op_ioepoch = md.body->ioepoch;
1358 ll_update_inode(inode, &md);
1359 ptlrpc_req_finished(request);
1364 /* Close IO epoch and send Size-on-MDS attribute update. */
1365 static int ll_setattr_done_writing(struct inode *inode,
1366 struct md_op_data *op_data,
1367 struct md_open_data *mod)
1369 struct ll_inode_info *lli = ll_i2info(inode);
1373 LASSERT(op_data != NULL);
1374 if (!S_ISREG(inode->i_mode))
1377 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1378 op_data->op_ioepoch, PFID(&lli->lli_fid));
1380 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1381 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1382 if (rc == -EAGAIN) {
1383 /* MDS has instructed us to obtain Size-on-MDS attribute
1384 * from OSTs and send setattr to back to MDS. */
1385 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1386 op_data->op_ioepoch);
1388 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1394 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1396 struct ll_sb_info *sbi = ll_i2sbi(inode);
1397 struct ll_inode_info *lli = ll_i2info(inode);
1398 struct lov_stripe_md *lsm = lli->lli_smd;
1400 ldlm_policy_data_t policy = { .l_extent = {new_size,
1402 struct lustre_handle lockh = { 0 };
1403 int local_lock = 0; /* 0 - no local lock;
1404 * 1 - lock taken by lock_extent;
1405 * 2 - by obd_match*/
1410 UNLOCK_INODE_MUTEX(inode);
1411 UP_WRITE_I_ALLOC_SEM(inode);
1413 if (sbi->ll_lockless_truncate_enable &&
1414 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1415 ast_flags = LDLM_FL_BLOCK_GRANTED;
1416 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1417 &policy, LCK_PW, &ast_flags, inode, &lockh);
1421 } else if (rc == 0) {
1422 rc = ll_file_punch(inode, new_size, 1);
1425 /* XXX when we fix the AST intents to pass the discard-range
1426 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1428 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1429 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1431 if (likely(rc == 0))
1435 LOCK_INODE_MUTEX(inode);
1436 DOWN_WRITE_I_ALLOC_SEM(inode);
1438 if (likely(rc == 0)) {
1439 /* Only ll_inode_size_lock is taken at this level.
1440 * lov_stripe_lock() is grabbed by ll_truncate() only over
1441 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1442 * ll_truncate dropped ll_inode_size_lock() */
1443 ll_inode_size_lock(inode, 0);
1445 spin_lock(&lli->lli_lock);
1446 lli->lli_flags |= LLIF_SRVLOCK;
1447 spin_unlock(&lli->lli_lock);
1449 rc = vmtruncate(inode, new_size);
1451 spin_lock(&lli->lli_lock);
1452 lli->lli_flags &= ~LLIF_SRVLOCK;
1453 spin_unlock(&lli->lli_lock);
1456 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1457 ll_inode_size_unlock(inode, 0);
1462 if (local_lock == 2)
1463 err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1465 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1466 if (unlikely(err != 0)){
1467 CERROR("extent unlock failed: err=%d,"
1468 " unlock method =%d\n", err, local_lock);
1476 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1477 * object(s) determine the file size and mtime. Otherwise, the MDS will
1478 * keep these values until such a time that objects are allocated for it.
1479 * We do the MDS operations first, as it is checking permissions for us.
1480 * We don't to the MDS RPC if there is nothing that we want to store there,
1481 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1482 * going to do an RPC anyways.
1484 * If we are doing a truncate, we will send the mtime and ctime updates
1485 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1486 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1489 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1491 struct ll_inode_info *lli = ll_i2info(inode);
1492 struct lov_stripe_md *lsm = lli->lli_smd;
1493 struct ll_sb_info *sbi = ll_i2sbi(inode);
1494 struct md_op_data *op_data = NULL;
1495 struct md_open_data *mod = NULL;
1496 int ia_valid = attr->ia_valid;
1497 int rc = 0, rc1 = 0;
1500 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1502 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1504 if (ia_valid & ATTR_SIZE) {
1505 if (attr->ia_size > ll_file_maxbytes(inode)) {
1506 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1507 attr->ia_size, ll_file_maxbytes(inode));
1511 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1514 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1515 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1516 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1520 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1521 if (attr->ia_valid & ATTR_CTIME) {
1522 attr->ia_ctime = CURRENT_TIME;
1523 attr->ia_valid |= ATTR_CTIME_SET;
1525 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1526 attr->ia_atime = CURRENT_TIME;
1527 attr->ia_valid |= ATTR_ATIME_SET;
1529 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1530 attr->ia_mtime = CURRENT_TIME;
1531 attr->ia_valid |= ATTR_MTIME_SET;
1533 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1534 /* To avoid stale mtime on mds, obtain it from ost and send
1536 rc = ll_glimpse_size(inode, 0);
1540 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1541 attr->ia_mtime = inode->i_mtime;
1544 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1545 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1546 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1547 cfs_time_current_sec());
1549 /* NB: ATTR_SIZE will only be set after this point if the size
1550 * resides on the MDS, ie, this file has no objects. */
1552 attr->ia_valid &= ~ATTR_SIZE;
1554 /* We always do an MDS RPC, even if we're only changing the size;
1555 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1557 OBD_ALLOC_PTR(op_data);
1558 if (op_data == NULL)
1561 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1563 /* Open epoch for truncate. */
1564 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1565 (ia_valid & ATTR_SIZE))
1566 op_data->op_flags = MF_EPOCH_OPEN;
1568 rc = ll_md_setattr(inode, op_data, &mod);
1572 if (op_data->op_ioepoch)
1573 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1574 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1576 if (!lsm || !S_ISREG(inode->i_mode)) {
1577 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1581 /* We really need to get our PW lock before we change inode->i_size.
1582 * If we don't we can race with other i_size updaters on our node, like
1583 * ll_file_read. We can also race with i_size propogation to other
1584 * nodes through dirtying and writeback of final cached pages. This
1585 * last one is especially bad for racing o_append users on other
1587 if (ia_valid & ATTR_SIZE) {
1588 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1589 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1591 struct obd_info oinfo = { { { 0 } } };
1594 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1595 inode->i_ino, LTIME_S(attr->ia_mtime));
1599 oa->o_id = lsm->lsm_object_id;
1600 oa->o_gr = lsm->lsm_object_gr;
1601 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1603 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1604 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1605 OBD_MD_FLFID | OBD_MD_FLGENER |
1608 obdo_from_inode(oa, inode, flags);
1612 oinfo.oi_capa = ll_mdscapa_get(inode);
1614 /* XXX: this looks unnecessary now. */
1615 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1616 capa_put(oinfo.oi_capa);
1618 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1627 if (op_data->op_ioepoch)
1628 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1629 ll_finish_md_op_data(op_data);
1631 return rc ? rc : rc1;
1634 int ll_setattr(struct dentry *de, struct iattr *attr)
1636 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1637 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1638 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1640 if ((de->d_inode->i_mode & S_ISUID) &&
1641 !(attr->ia_mode & S_ISUID) &&
1642 !(attr->ia_valid & ATTR_KILL_SUID))
1643 attr->ia_valid |= ATTR_KILL_SUID;
1645 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1646 !(attr->ia_mode & S_ISGID) &&
1647 !(attr->ia_valid & ATTR_KILL_SGID))
1648 attr->ia_valid |= ATTR_KILL_SGID;
1650 return ll_setattr_raw(de->d_inode, attr);
1653 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1654 __u64 max_age, __u32 flags)
1656 struct ll_sb_info *sbi = ll_s2sbi(sb);
1657 struct obd_statfs obd_osfs;
1661 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1663 CERROR("md_statfs fails: rc = %d\n", rc);
1667 osfs->os_type = sb->s_magic;
1669 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1670 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1672 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1673 &obd_osfs, max_age, flags);
1675 CERROR("obd_statfs fails: rc = %d\n", rc);
1679 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1680 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1683 osfs->os_bsize = obd_osfs.os_bsize;
1684 osfs->os_blocks = obd_osfs.os_blocks;
1685 osfs->os_bfree = obd_osfs.os_bfree;
1686 osfs->os_bavail = obd_osfs.os_bavail;
1688 /* If we don't have as many objects free on the OST as inodes
1689 * on the MDS, we reduce the total number of inodes to
1690 * compensate, so that the "inodes in use" number is correct.
1692 if (obd_osfs.os_ffree < osfs->os_ffree) {
1693 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1695 osfs->os_ffree = obd_osfs.os_ffree;
1700 #ifndef HAVE_STATFS_DENTRY_PARAM
1701 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1704 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1706 struct super_block *sb = de->d_sb;
1708 struct obd_statfs osfs;
1711 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1712 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1714 /* For now we will always get up-to-date statfs values, but in the
1715 * future we may allow some amount of caching on the client (e.g.
1716 * from QOS or lprocfs updates). */
1717 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1721 statfs_unpack(sfs, &osfs);
1723 /* We need to downshift for all 32-bit kernels, because we can't
1724 * tell if the kernel is being called via sys_statfs64() or not.
1725 * Stop before overflowing f_bsize - in which case it is better
1726 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1727 if (sizeof(long) < 8) {
1728 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1731 osfs.os_blocks >>= 1;
1732 osfs.os_bfree >>= 1;
1733 osfs.os_bavail >>= 1;
1737 sfs->f_blocks = osfs.os_blocks;
1738 sfs->f_bfree = osfs.os_bfree;
1739 sfs->f_bavail = osfs.os_bavail;
1744 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1746 struct ll_inode_info *lli;
1747 struct lov_stripe_md *lsm;
1749 lli = ll_i2info(inode);
1750 LASSERT(lli->lli_size_sem_owner != current);
1751 down(&lli->lli_size_sem);
1752 LASSERT(lli->lli_size_sem_owner == NULL);
1753 lli->lli_size_sem_owner = current;
1755 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1758 lov_stripe_lock(lsm);
1761 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1763 struct ll_inode_info *lli;
1764 struct lov_stripe_md *lsm;
1766 lli = ll_i2info(inode);
1768 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1771 lov_stripe_unlock(lsm);
1772 LASSERT(lli->lli_size_sem_owner == current);
1773 lli->lli_size_sem_owner = NULL;
1774 up(&lli->lli_size_sem);
1777 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1779 struct ll_inode_info *lli = ll_i2info(inode);
1781 dump_lsm(D_INODE, lsm);
1782 dump_lsm(D_INODE, lli->lli_smd);
1783 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1784 "lsm must be joined lsm %p\n", lsm);
1785 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1786 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1787 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1789 lli->lli_maxbytes = lsm->lsm_maxbytes;
1790 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1791 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1794 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1796 struct ll_inode_info *lli = ll_i2info(inode);
1797 struct mdt_body *body = md->body;
1798 struct lov_stripe_md *lsm = md->lsm;
1799 struct ll_sb_info *sbi = ll_i2sbi(inode);
1801 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1803 if (lli->lli_smd == NULL) {
1804 if (lsm->lsm_magic != LOV_MAGIC &&
1805 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1806 dump_lsm(D_ERROR, lsm);
1809 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1810 lsm, inode->i_ino, inode->i_generation, inode);
1811 /* ll_inode_size_lock() requires it is only called
1812 * with lli_smd != NULL or lock_lsm == 0 or we can
1813 * race between lock/unlock. bug 9547 */
1815 lli->lli_maxbytes = lsm->lsm_maxbytes;
1816 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1817 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1819 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1820 lli->lli_smd->lsm_stripe_count ==
1821 lsm->lsm_stripe_count) {
1822 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1823 CERROR("lsm mismatch for inode %ld\n",
1825 CERROR("lli_smd:\n");
1826 dump_lsm(D_ERROR, lli->lli_smd);
1828 dump_lsm(D_ERROR, lsm);
1832 ll_replace_lsm(inode, lsm);
1834 if (lli->lli_smd != lsm)
1835 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1838 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1839 if (body->valid & OBD_MD_FLRMTPERM)
1840 ll_update_remote_perm(inode, md->remote_perm);
1842 #ifdef CONFIG_FS_POSIX_ACL
1843 else if (body->valid & OBD_MD_FLACL) {
1844 spin_lock(&lli->lli_lock);
1845 if (lli->lli_posix_acl)
1846 posix_acl_release(lli->lli_posix_acl);
1847 lli->lli_posix_acl = md->posix_acl;
1848 spin_unlock(&lli->lli_lock);
1851 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1853 if (body->valid & OBD_MD_FLATIME &&
1854 body->atime > LTIME_S(inode->i_atime))
1855 LTIME_S(inode->i_atime) = body->atime;
1857 /* mtime is always updated with ctime, but can be set in past.
1858 As write and utime(2) may happen within 1 second, and utime's
1859 mtime has a priority over write's one, so take mtime from mds
1860 for the same ctimes. */
1861 if (body->valid & OBD_MD_FLCTIME &&
1862 body->ctime >= LTIME_S(inode->i_ctime)) {
1863 LTIME_S(inode->i_ctime) = body->ctime;
1864 if (body->valid & OBD_MD_FLMTIME) {
1865 CDEBUG(D_INODE, "setting ino %lu mtime "
1866 "from %lu to "LPU64"\n", inode->i_ino,
1867 LTIME_S(inode->i_mtime), body->mtime);
1868 LTIME_S(inode->i_mtime) = body->mtime;
1871 if (body->valid & OBD_MD_FLMODE)
1872 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1873 if (body->valid & OBD_MD_FLTYPE)
1874 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1875 if (S_ISREG(inode->i_mode)) {
1876 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1878 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1880 #ifdef HAVE_INODE_BLKSIZE
1881 inode->i_blksize = 1<<inode->i_blkbits;
1883 if (body->valid & OBD_MD_FLUID)
1884 inode->i_uid = body->uid;
1885 if (body->valid & OBD_MD_FLGID)
1886 inode->i_gid = body->gid;
1887 if (body->valid & OBD_MD_FLFLAGS)
1888 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1889 if (body->valid & OBD_MD_FLNLINK)
1890 inode->i_nlink = body->nlink;
1891 if (body->valid & OBD_MD_FLRDEV)
1892 inode->i_rdev = old_decode_dev(body->rdev);
1894 if (body->valid & OBD_MD_FLID) {
1895 /* FID shouldn't be changed! */
1896 if (fid_is_sane(&lli->lli_fid)) {
1897 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1898 "Trying to change FID "DFID
1899 " to the "DFID", inode %lu/%u(%p)\n",
1900 PFID(&lli->lli_fid), PFID(&body->fid1),
1901 inode->i_ino, inode->i_generation, inode);
1903 lli->lli_fid = body->fid1;
1906 LASSERT(fid_seq(&lli->lli_fid) != 0);
1908 if (body->valid & OBD_MD_FLSIZE) {
1909 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1910 S_ISREG(inode->i_mode) && lli->lli_smd) {
1911 struct lustre_handle lockh;
1914 /* As it is possible a blocking ast has been processed
1915 * by this time, we need to check there is an UPDATE
1916 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1918 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1921 if (lli->lli_flags & (LLIF_DONE_WRITING |
1922 LLIF_EPOCH_PENDING |
1924 CERROR("ino %lu flags %lu still has "
1925 "size authority! do not trust "
1926 "the size got from MDS\n",
1927 inode->i_ino, lli->lli_flags);
1929 /* Use old size assignment to avoid
1930 * deadlock bz14138 & bz14326 */
1931 inode->i_size = body->size;
1932 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1934 ldlm_lock_decref(&lockh, mode);
1937 /* Use old size assignment to avoid
1938 * deadlock bz14138 & bz14326 */
1939 inode->i_size = body->size;
1942 if (body->valid & OBD_MD_FLBLOCKS)
1943 inode->i_blocks = body->blocks;
1946 if (body->valid & OBD_MD_FLMDSCAPA) {
1947 LASSERT(md->mds_capa);
1948 ll_add_capa(inode, md->mds_capa);
1950 if (body->valid & OBD_MD_FLOSSCAPA) {
1951 LASSERT(md->oss_capa);
1952 ll_add_capa(inode, md->oss_capa);
1956 static struct backing_dev_info ll_backing_dev_info = {
1957 .ra_pages = 0, /* No readahead */
1958 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1959 .capabilities = 0, /* Does contribute to dirty memory */
1961 .memory_backed = 0, /* Does contribute to dirty memory */
1965 void ll_read_inode2(struct inode *inode, void *opaque)
1967 struct lustre_md *md = opaque;
1968 struct ll_inode_info *lli = ll_i2info(inode);
1971 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1972 inode->i_ino, inode->i_generation, inode);
1976 LASSERT(!lli->lli_smd);
1978 /* Core attributes from the MDS first. This is a new inode, and
1979 * the VFS doesn't zero times in the core inode so we have to do
1980 * it ourselves. They will be overwritten by either MDS or OST
1981 * attributes - we just need to make sure they aren't newer. */
1982 LTIME_S(inode->i_mtime) = 0;
1983 LTIME_S(inode->i_atime) = 0;
1984 LTIME_S(inode->i_ctime) = 0;
1986 ll_update_inode(inode, md);
1988 /* OIDEBUG(inode); */
1990 if (S_ISREG(inode->i_mode)) {
1991 struct ll_sb_info *sbi = ll_i2sbi(inode);
1992 inode->i_op = &ll_file_inode_operations;
1993 inode->i_fop = sbi->ll_fop;
1994 inode->i_mapping->a_ops = &ll_aops;
1996 } else if (S_ISDIR(inode->i_mode)) {
1997 inode->i_op = &ll_dir_inode_operations;
1998 inode->i_fop = &ll_dir_operations;
1999 inode->i_mapping->a_ops = &ll_dir_aops;
2001 } else if (S_ISLNK(inode->i_mode)) {
2002 inode->i_op = &ll_fast_symlink_inode_operations;
2005 inode->i_op = &ll_special_inode_operations;
2007 init_special_inode(inode, inode->i_mode,
2008 kdev_t_to_nr(inode->i_rdev));
2010 /* initializing backing dev info. */
2011 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2017 void ll_delete_inode(struct inode *inode)
2019 struct ll_sb_info *sbi = ll_i2sbi(inode);
2023 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2025 CERROR("fid_delete() failed, rc %d\n", rc);
2027 truncate_inode_pages(&inode->i_data, 0);
2033 int ll_iocontrol(struct inode *inode, struct file *file,
2034 unsigned int cmd, unsigned long arg)
2036 struct ll_sb_info *sbi = ll_i2sbi(inode);
2037 struct ptlrpc_request *req = NULL;
2042 case EXT3_IOC_GETFLAGS: {
2043 struct mdt_body *body;
2044 struct obd_capa *oc;
2046 oc = ll_mdscapa_get(inode);
2047 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2048 OBD_MD_FLFLAGS, 0, &req);
2051 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2055 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2057 flags = body->flags;
2059 ptlrpc_req_finished(req);
2061 RETURN(put_user(flags, (int *)arg));
2063 case EXT3_IOC_SETFLAGS: {
2064 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2065 struct obd_info oinfo = { { { 0 } } };
2066 struct md_op_data *op_data;
2068 if (get_user(flags, (int *)arg))
2072 OBDO_ALLOC(oinfo.oi_oa);
2076 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2077 LUSTRE_OPC_ANY, NULL);
2078 if (IS_ERR(op_data))
2079 RETURN(PTR_ERR(op_data));
2081 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2082 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2083 rc = md_setattr(sbi->ll_md_exp, op_data,
2084 NULL, 0, NULL, 0, &req, NULL);
2085 ll_finish_md_op_data(op_data);
2086 ptlrpc_req_finished(req);
2087 if (rc || lsm == NULL) {
2088 OBDO_FREE(oinfo.oi_oa);
2092 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2093 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2094 oinfo.oi_oa->o_flags = flags;
2095 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2097 oinfo.oi_capa = ll_mdscapa_get(inode);
2099 obdo_from_inode(oinfo.oi_oa, inode,
2100 OBD_MD_FLFID | OBD_MD_FLGENER);
2101 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2102 capa_put(oinfo.oi_capa);
2103 OBDO_FREE(oinfo.oi_oa);
2105 if (rc != -EPERM && rc != -EACCES)
2106 CERROR("md_setattr_async fails: rc = %d\n", rc);
2110 inode->i_flags = ll_ext_to_inode_flags(flags |
2111 MDS_BFLAG_EXT_FLAGS);
2121 int ll_flush_ctx(struct inode *inode)
2123 struct ll_sb_info *sbi = ll_i2sbi(inode);
2125 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2127 obd_set_info_async(sbi->ll_md_exp,
2128 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2130 obd_set_info_async(sbi->ll_dt_exp,
2131 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2136 /* umount -f client means force down, don't save state */
2137 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2138 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2140 struct super_block *sb = vfsmnt->mnt_sb;
2142 void ll_umount_begin(struct super_block *sb)
2145 struct lustre_sb_info *lsi = s2lsi(sb);
2146 struct ll_sb_info *sbi = ll_s2sbi(sb);
2147 struct obd_device *obd;
2148 struct obd_ioctl_data ioc_data = { 0 };
2151 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2152 if (!(flags & MNT_FORCE)) {
2158 /* Tell the MGC we got umount -f */
2159 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2161 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2162 sb->s_count, atomic_read(&sb->s_active));
2164 obd = class_exp2obd(sbi->ll_md_exp);
2166 CERROR("Invalid MDC connection handle "LPX64"\n",
2167 sbi->ll_md_exp->exp_handle.h_cookie);
2172 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2175 obd = class_exp2obd(sbi->ll_dt_exp);
2177 CERROR("Invalid LOV connection handle "LPX64"\n",
2178 sbi->ll_dt_exp->exp_handle.h_cookie);
2184 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2187 /* Really, we'd like to wait until there are no requests outstanding,
2188 * and then continue. For now, we just invalidate the requests,
2189 * schedule, and hope.
2196 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2198 struct ll_sb_info *sbi = ll_s2sbi(sb);
2202 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2203 read_only = *flags & MS_RDONLY;
2204 err = obd_set_info_async(sbi->ll_md_exp,
2205 sizeof(KEY_READ_ONLY),
2206 KEY_READ_ONLY, sizeof(read_only),
2209 CERROR("Failed to change the read-only flag during "
2210 "remount: %d\n", err);
2215 sb->s_flags |= MS_RDONLY;
2217 sb->s_flags &= ~MS_RDONLY;
2222 int ll_prep_inode(struct inode **inode,
2223 struct ptlrpc_request *req,
2224 struct super_block *sb)
2226 struct ll_sb_info *sbi = NULL;
2227 struct lustre_md md;
2231 LASSERT(*inode || sb);
2232 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2233 prune_deathrow(sbi, 1);
2234 memset(&md, 0, sizeof(struct lustre_md));
2236 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2237 sbi->ll_md_exp, &md);
2242 ll_update_inode(*inode, &md);
2244 LASSERT(sb != NULL);
2247 * At this point server returns to client's same fid as client
2248 * generated for creating. So using ->fid1 is okay here.
2250 LASSERT(fid_is_sane(&md.body->fid1));
2252 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2253 if (*inode == NULL || is_bad_inode(*inode)) {
2255 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2256 #ifdef CONFIG_FS_POSIX_ACL
2258 posix_acl_release(md.posix_acl);
2259 md.posix_acl = NULL;
2263 CERROR("new_inode -fatal: rc %d\n", rc);
2268 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2269 ll_i2info(*inode)->lli_smd);
2271 md_free_lustre_md(sbi->ll_md_exp, &md);
2275 char *llap_origins[] = {
2276 [LLAP_ORIGIN_UNKNOWN] = "--",
2277 [LLAP_ORIGIN_READPAGE] = "rp",
2278 [LLAP_ORIGIN_READAHEAD] = "ra",
2279 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2280 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2281 [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2284 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2285 struct list_head *list)
2287 struct ll_async_page *llap;
2288 struct list_head *pos;
2290 list_for_each(pos, list) {
2293 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2294 if (llap->llap_page == NULL)
2302 int ll_obd_statfs(struct inode *inode, void *arg)
2304 struct ll_sb_info *sbi = NULL;
2305 struct obd_export *exp;
2307 struct obd_ioctl_data *data = NULL;
2311 if (!inode || !(sbi = ll_i2sbi(inode)))
2312 GOTO(out_statfs, rc = -EINVAL);
2314 rc = obd_ioctl_getdata(&buf, &len, arg);
2316 GOTO(out_statfs, rc);
2319 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2320 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2321 GOTO(out_statfs, rc = -EINVAL);
2323 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2324 if (type == LL_STATFS_MDC)
2325 exp = sbi->ll_md_exp;
2326 else if (type == LL_STATFS_LOV)
2327 exp = sbi->ll_dt_exp;
2329 GOTO(out_statfs, rc = -ENODEV);
2331 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2333 GOTO(out_statfs, rc);
2336 obd_ioctl_freedata(buf, len);
2340 int ll_process_config(struct lustre_cfg *lcfg)
2344 struct lprocfs_static_vars lvars;
2348 lprocfs_llite_init_vars(&lvars);
2350 /* The instance name contains the sb: lustre-client-aacfe000 */
2351 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2352 if (!ptr || !*(++ptr))
2354 if (sscanf(ptr, "%lx", &x) != 1)
2357 /* This better be a real Lustre superblock! */
2358 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2360 /* Note we have not called client_common_fill_super yet, so
2361 proc fns must be able to handle that! */
2362 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2367 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2368 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2369 struct inode *i1, struct inode *i2,
2370 const char *name, int namelen,
2371 int mode, __u32 opc, void *data)
2373 LASSERT(i1 != NULL);
2375 if (namelen > ll_i2sbi(i1)->ll_namelen)
2376 return ERR_PTR(-ENAMETOOLONG);
2378 if (op_data == NULL)
2379 OBD_ALLOC_PTR(op_data);
2381 if (op_data == NULL)
2382 return ERR_PTR(-ENOMEM);
2384 ll_i2gids(op_data->op_suppgids, i1, i2);
2385 op_data->op_fid1 = *ll_inode2fid(i1);
2386 op_data->op_capa1 = ll_mdscapa_get(i1);
2389 op_data->op_fid2 = *ll_inode2fid(i2);
2390 op_data->op_capa2 = ll_mdscapa_get(i2);
2392 fid_zero(&op_data->op_fid2);
2393 op_data->op_capa2 = NULL;
2396 op_data->op_name = name;
2397 op_data->op_namelen = namelen;
2398 op_data->op_mode = mode;
2399 op_data->op_mod_time = cfs_time_current_sec();
2400 op_data->op_fsuid = current->fsuid;
2401 op_data->op_fsgid = current->fsgid;
2402 op_data->op_cap = current->cap_effective;
2403 op_data->op_bias = MDS_CHECK_SPLIT;
2404 op_data->op_opc = opc;
2405 op_data->op_mds = 0;
2406 op_data->op_data = data;
2411 void ll_finish_md_op_data(struct md_op_data *op_data)
2413 capa_put(op_data->op_capa1);
2414 capa_put(op_data->op_capa2);
2415 OBD_FREE_PTR(op_data);
2418 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2420 struct ll_sb_info *sbi;
2422 LASSERT((seq != NULL) && (vfs != NULL));
2423 sbi = ll_s2sbi(vfs->mnt_sb);
2425 if (sbi->ll_flags & LL_SBI_NOLCK)
2426 seq_puts(seq, ",nolock");
2428 if (sbi->ll_flags & LL_SBI_FLOCK)
2429 seq_puts(seq, ",flock");
2431 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2432 seq_puts(seq, ",localflock");
2434 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2435 seq_puts(seq, ",user_xattr");
2437 if (sbi->ll_flags & LL_SBI_ACL)
2438 seq_puts(seq, ",acl");