4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Directory code for lustre client.
38 #include <linux/pagemap.h>
40 #include <linux/version.h>
41 #include <linux/security.h>
42 #include <linux/user_namespace.h>
43 #ifdef HAVE_UIDGID_HEADER
44 # include <linux/uidgid.h>
46 #include <linux/uaccess.h>
47 #include <linux/buffer_head.h> // for wait_on_buffer
48 #include <linux/pagevec.h>
50 #define DEBUG_SUBSYSTEM S_LLITE
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #include <lustre_lib.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include <lustre_kernelcomm.h>
59 #include <lustre_swab.h>
61 #include "llite_internal.h"
64 * (new) readdir implementation overview.
66 * Original lustre readdir implementation cached exact copy of raw directory
67 * pages on the client. These pages were indexed in client page cache by
68 * logical offset in the directory file. This design, while very simple and
69 * intuitive had some inherent problems:
71 * . it implies that byte offset to the directory entry serves as a
72 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
73 * ext3/htree directory entries may move due to splits, and more
76 * . it is incompatible with the design of split directories for cmd3,
77 * that assumes that names are distributed across nodes based on their
78 * hash, and so readdir should be done in hash order.
80 * New readdir implementation does readdir in hash order, and uses hash of a
81 * file name as a telldir/seekdir cookie. This led to number of complications:
83 * . hash is not unique, so it cannot be used to index cached directory
84 * pages on the client (note, that it requires a whole pageful of hash
85 * collided entries to cause two pages to have identical hashes);
87 * . hash is not unique, so it cannot, strictly speaking, be used as an
88 * entry cookie. ext3/htree has the same problem and lustre implementation
89 * mimics their solution: seekdir(hash) positions directory at the first
90 * entry with the given hash.
96 * Client caches directory pages using hash of the first entry as an index. As
97 * noted above hash is not unique, so this solution doesn't work as is:
98 * special processing is needed for "page hash chains" (i.e., sequences of
99 * pages filled with entries all having the same hash value).
101 * First, such chains have to be detected. To this end, server returns to the
102 * client the hash of the first entry on the page next to one returned. When
103 * client detects that this hash is the same as hash of the first entry on the
104 * returned page, page hash collision has to be handled. Pages in the
105 * hash chain, except first one, are termed "overflow pages".
107 * Solution to index uniqueness problem is to not cache overflow
108 * pages. Instead, when page hash collision is detected, all overflow pages
109 * from emerging chain are immediately requested from the server and placed in
110 * a special data structure (struct ll_dir_chain). This data structure is used
111 * by ll_readdir() to process entries from overflow pages. When readdir
112 * invocation finishes, overflow pages are discarded. If page hash collision
113 * chain weren't completely processed, next call to readdir will again detect
114 * page hash collision, again read overflow pages in, process next portion of
115 * entries and again discard the pages. This is not as wasteful as it looks,
116 * because, given reasonable hash, page hash collisions are extremely rare.
118 * 1. directory positioning
120 * When seekdir(hash) is called, original
131 * identification of and access to overflow pages
135 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
136 * a header lu_dirpage which describes the start/end hash, and whether this
137 * page is empty (contains no dir entry) or hash collide with next page.
138 * After client receives reply, several pages will be integrated into dir page
139 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
140 * lu_dirpage for this integrated page will be adjusted. See
141 * mdc_adjust_dirpages().
144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
145 __u64 offset, struct ll_dir_chain *chain)
147 struct md_callback cb_op;
151 cb_op.md_blocking_ast = ll_md_blocking_ast;
152 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
159 void ll_release_page(struct inode *inode, struct page *page,
164 /* Always remove the page for striped dir, because the page is
165 * built from temporarily in LMV layer */
166 if (inode != NULL && S_ISDIR(inode->i_mode) &&
167 ll_i2info(inode)->lli_lsm_md != NULL) {
174 if (likely(page->mapping != NULL))
175 truncate_complete_page(page->mapping, page);
182 * return IF_* type for given lu_dirent entry.
183 * IF_* flag shld be converted to particular OS file type in
184 * platform llite module.
186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
189 struct luda_type *lt;
192 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
193 const unsigned align = sizeof(struct luda_type) - 1;
195 len = le16_to_cpu(ent->lde_namelen);
196 len = (len + align) & ~align;
197 lt = (void *)ent->lde_name + len;
198 type = IFTODT(le16_to_cpu(lt->lt_type));
204 #ifdef HAVE_DIR_CONTEXT
205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
206 struct dir_context *ctx)
209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
210 void *cookie, filldir_t filldir)
213 struct ll_sb_info *sbi = ll_i2sbi(inode);
215 bool is_api32 = ll_need_32bit_api(sbi);
216 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
218 struct ll_dir_chain chain;
223 ll_dir_chain_init(&chain);
225 page = ll_get_dir_page(inode, op_data, pos, &chain);
227 while (rc == 0 && !done) {
228 struct lu_dirpage *dp;
229 struct lu_dirent *ent;
238 hash = MDS_DIR_END_OFF;
239 dp = page_address(page);
240 for (ent = lu_dirent_start(dp); ent != NULL && !done;
241 ent = lu_dirent_next(ent)) {
248 hash = le64_to_cpu(ent->lde_hash);
249 if (hash < pos) /* Skip until we find target hash */
252 namelen = le16_to_cpu(ent->lde_namelen);
253 if (namelen == 0) /* Skip dummy record */
256 if (is_api32 && is_hash64)
260 fid_le_to_cpu(&fid, &ent->lde_fid);
261 ino = cl_fid_build_ino(&fid, is_api32);
262 type = ll_dirent_type_get(ent);
263 /* For ll_nfs_get_name_filldir(), it will try to access
264 * 'ent' through 'lde_name', so the parameter 'name'
265 * for 'filldir()' must be part of the 'ent'. */
266 #ifdef HAVE_DIR_CONTEXT
268 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
271 done = filldir(cookie, ent->lde_name, namelen, lhash,
278 ll_release_page(inode, page, false);
282 next = le64_to_cpu(dp->ldp_hash_end);
284 if (pos == MDS_DIR_END_OFF) {
286 * End of directory reached.
289 ll_release_page(inode, page, false);
292 * Normal case: continue to the next
295 ll_release_page(inode, page,
296 le32_to_cpu(dp->ldp_flags) &
299 page = ll_get_dir_page(inode, op_data, pos,
303 #ifdef HAVE_DIR_CONTEXT
308 ll_dir_chain_fini(&chain);
312 #ifdef HAVE_DIR_CONTEXT
313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
318 struct inode *inode = file_inode(filp);
319 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
320 struct ll_sb_info *sbi = ll_i2sbi(inode);
321 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
322 int api32 = ll_need_32bit_api(sbi);
323 struct md_op_data *op_data;
324 struct lu_fid pfid = { 0 };
334 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
335 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
336 inode, (unsigned long)pos, i_size_read(inode), api32);
338 if (pos == MDS_DIR_END_OFF)
344 if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
345 /* This is only needed for striped dir to fill ..,
346 * see lmv_read_entry */
347 if (file_dentry(filp)->d_parent != NULL &&
348 file_dentry(filp)->d_parent->d_inode != NULL) {
349 __u64 ibits = MDS_INODELOCK_UPDATE;
350 struct inode *parent =
351 file_dentry(filp)->d_parent->d_inode;
353 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
354 pfid = *ll_inode2fid(parent);
357 /* If it can not find in cache, do lookup .. on the master
359 if (fid_is_zero(&pfid)) {
360 rc = ll_dir_get_parent_fid(inode, &pfid);
366 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
367 LUSTRE_OPC_ANY, inode);
369 GOTO(out, rc = PTR_ERR(op_data));
370 op_data->op_fid3 = pfid;
372 #ifdef HAVE_DIR_CONTEXT
374 rc = ll_dir_read(inode, &pos, op_data, ctx);
377 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
382 if (pos == MDS_DIR_END_OFF) {
384 pos = LL_DIR_END_OFF_32BIT;
386 pos = LL_DIR_END_OFF;
391 #ifdef HAVE_DIR_CONTEXT
396 ll_finish_md_op_data(op_data);
400 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
405 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
406 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
408 struct mgs_send_param *msp;
415 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
416 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
417 sizeof(struct mgs_send_param), msp, NULL);
419 CERROR("Failed to set parameter: %d\n", rc);
427 * Create striped directory with specified stripe(@lump)
429 * \param[in] dparent the parent of the directory.
430 * \param[in] lump the specified stripes.
431 * \param[in] dirname the name of the directory.
432 * \param[in] mode the specified mode of the directory.
434 * \retval =0 if striped directory is being created successfully.
435 * <0 if the creation is failed.
437 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
438 size_t len, const char *dirname, umode_t mode)
440 struct inode *parent = dparent->d_inode;
441 struct ptlrpc_request *request = NULL;
442 struct md_op_data *op_data;
443 struct ll_sb_info *sbi = ll_i2sbi(parent);
444 struct inode *inode = NULL;
445 struct dentry dentry = {
449 .len = strlen(dirname),
450 .hash = ll_full_name_hash(dparent, dirname,
457 if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
458 lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
461 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
462 "stripe_offset %d, stripe_count: %u\n",
463 PFID(ll_inode2fid(parent)), parent, dirname,
464 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
466 if (lump->lum_stripe_count > 1 &&
467 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
470 if (IS_DEADDIR(parent) &&
471 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
474 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
475 lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
476 lustre_swab_lmv_user_md(lump);
478 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
479 mode &= ~current_umask();
480 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
481 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
482 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
485 RETURN(PTR_ERR(op_data));
487 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
488 /* selinux_dentry_init_security() uses dentry->d_parent and name
489 * to determine the security context for the file. So our fake
490 * dentry should be real enough for this purpose. */
491 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
492 &op_data->op_file_secctx_name,
493 &op_data->op_file_secctx,
494 &op_data->op_file_secctx_size);
496 GOTO(out_op_data, err);
499 op_data->op_cli_flags |= CLI_SET_MEA;
500 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
501 from_kuid(&init_user_ns, current_fsuid()),
502 from_kgid(&init_user_ns, current_fsgid()),
503 cfs_curproc_cap_pack(), 0, &request);
505 GOTO(out_request, err);
507 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
509 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
511 GOTO(out_inode, err);
513 dentry.d_inode = inode;
515 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
517 err = security_inode_notifysecctx(inode,
518 op_data->op_file_secctx,
519 op_data->op_file_secctx_size);
522 err = ll_inode_init_security(&dentry, inode, parent);
525 GOTO(out_inode, err);
531 ptlrpc_req_finished(request);
533 ll_finish_md_op_data(op_data);
538 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
541 struct ll_sb_info *sbi = ll_i2sbi(inode);
542 struct md_op_data *op_data;
543 struct ptlrpc_request *req = NULL;
545 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
546 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
547 struct obd_device *mgc = lsi->lsi_mgc;
554 * This is coming from userspace, so should be in
555 * local endian. But the MDS would like it in little
556 * endian, so we swab it before we send it.
558 switch (lump->lmm_magic) {
559 case LOV_USER_MAGIC_V1: {
560 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
561 lustre_swab_lov_user_md_v1(lump);
562 lum_size = sizeof(struct lov_user_md_v1);
565 case LOV_USER_MAGIC_V3: {
566 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
567 lustre_swab_lov_user_md_v3(
568 (struct lov_user_md_v3 *)lump);
569 lum_size = sizeof(struct lov_user_md_v3);
572 case LOV_USER_MAGIC_COMP_V1: {
573 if (lump->lmm_magic !=
574 cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
575 lustre_swab_lov_comp_md_v1(
576 (struct lov_comp_md_v1 *)lump);
577 lum_size = le32_to_cpu(
578 ((struct lov_comp_md_v1 *)lump)->lcm_size);
581 case LMV_USER_MAGIC: {
582 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
583 lustre_swab_lmv_user_md(
584 (struct lmv_user_md *)lump);
585 lum_size = sizeof(struct lmv_user_md);
588 case LOV_USER_MAGIC_SPECIFIC: {
589 struct lov_user_md_v3 *v3 =
590 (struct lov_user_md_v3 *)lump;
591 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
593 if (lump->lmm_magic !=
594 cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
595 lustre_swab_lov_user_md_v3(v3);
596 lustre_swab_lov_user_md_objects(v3->lmm_objects,
597 v3->lmm_stripe_count);
599 lum_size = lov_user_md_size(v3->lmm_stripe_count,
600 LOV_USER_MAGIC_SPECIFIC);
604 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
605 " %#08x != %#08x nor %#08x\n",
606 lump->lmm_magic, LOV_USER_MAGIC_V1,
612 lum_size = sizeof(struct lov_user_md_v1);
615 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
616 LUSTRE_OPC_ANY, NULL);
618 RETURN(PTR_ERR(op_data));
620 /* swabbing is done in lov_setstripe() on server side */
621 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
622 ll_finish_md_op_data(op_data);
623 ptlrpc_req_finished(req);
627 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
629 * 2.9 server has stored filesystem default stripe in ROOT xattr,
630 * and it's stored into system config for backward compatibility.
632 * In the following we use the fact that LOV_USER_MAGIC_V1 and
633 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
634 * need the make the distiction between the 2 versions
636 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
638 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
639 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
643 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
645 GOTO(end, rc = -ENOMEM);
648 /* Get fsname and assume devname to be -MDT0000. */
649 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
650 strcat(buf, "-MDT0000.lov");
653 /* Set root stripesize */
654 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
655 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
656 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
660 /* Set root stripecount */
661 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
662 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
663 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
667 /* Set root stripeoffset */
668 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
669 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
670 (typeof(lump->lmm_stripe_offset))(-1));
671 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
675 OBD_FREE(param, MGS_PARAM_MAXLEN);
682 * This function will be used to get default LOV/LMV/Default LMV
683 * @valid will be used to indicate which stripe it will retrieve
684 * OBD_MD_MEA LMV stripe EA
685 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
686 * otherwise Default LOV EA.
687 * Each time, it can only retrieve 1 stripe EA
689 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
690 struct ptlrpc_request **request, u64 valid)
692 struct ll_sb_info *sbi = ll_i2sbi(inode);
693 struct mdt_body *body;
694 struct lov_mds_md *lmm = NULL;
695 struct ptlrpc_request *req = NULL;
697 struct md_op_data *op_data;
700 rc = ll_get_default_mdsize(sbi, &lmm_size);
704 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
705 0, lmm_size, LUSTRE_OPC_ANY,
708 RETURN(PTR_ERR(op_data));
710 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
711 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
712 ll_finish_md_op_data(op_data);
714 CDEBUG(D_INFO, "md_getattr failed on inode "
715 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
719 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
720 LASSERT(body != NULL);
722 lmm_size = body->mbo_eadatasize;
724 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
726 GOTO(out, rc = -ENODATA);
729 lmm = req_capsule_server_sized_get(&req->rq_pill,
730 &RMF_MDT_MD, lmm_size);
731 LASSERT(lmm != NULL);
734 * This is coming from the MDS, so is probably in
735 * little endian. We convert it to host endian before
736 * passing it to userspace.
738 /* We don't swab objects for directories */
739 switch (le32_to_cpu(lmm->lmm_magic)) {
741 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
742 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
745 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
746 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
748 case LOV_MAGIC_COMP_V1:
749 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
750 lustre_swab_lov_comp_md_v1(
751 (struct lov_comp_md_v1 *)lmm);
754 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
755 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
758 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
759 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
761 case LOV_USER_MAGIC_SPECIFIC: {
762 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
764 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
765 lustre_swab_lov_user_md_v3(v3);
766 lustre_swab_lov_user_md_objects(v3->lmm_objects,
767 v3->lmm_stripe_count);
772 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
777 *plmm_size = lmm_size;
782 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
784 struct md_op_data *op_data;
789 OBD_ALLOC_PTR(op_data);
793 op_data->op_flags |= MF_GET_MDT_IDX;
794 op_data->op_fid1 = *fid;
795 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
796 mdt_index = op_data->op_mds;
797 OBD_FREE_PTR(op_data);
805 * Get MDT index for the inode.
807 int ll_get_mdt_idx(struct inode *inode)
809 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
813 * Generic handler to do any pre-copy work.
815 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
816 * first information for it that real work has started.
818 * Moreover, for a ARCHIVE request, it will sample the file data version and
819 * store it in \a copy.
821 * \return 0 on success.
823 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
825 struct ll_sb_info *sbi = ll_s2sbi(sb);
826 struct hsm_progress_kernel hpk;
831 /* Forge a hsm_progress based on data from copy. */
832 hpk.hpk_fid = copy->hc_hai.hai_fid;
833 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
834 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
835 hpk.hpk_extent.length = 0;
838 hpk.hpk_data_version = 0;
841 /* For archive request, we need to read the current file version. */
842 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
844 __u64 data_version = 0;
846 /* Get inode for this fid */
847 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
849 hpk.hpk_flags |= HP_FLAG_RETRY;
850 /* hpk_errval is >= 0 */
851 hpk.hpk_errval = -PTR_ERR(inode);
852 GOTO(progress, rc = PTR_ERR(inode));
855 /* Read current file data version */
856 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
859 CDEBUG(D_HSM, "Could not read file data version of "
860 DFID" (rc = %d). Archive request ("
861 "%#llx) could not be done.\n",
862 PFID(©->hc_hai.hai_fid), rc,
863 copy->hc_hai.hai_cookie);
864 hpk.hpk_flags |= HP_FLAG_RETRY;
865 /* hpk_errval must be >= 0 */
866 hpk.hpk_errval = -rc;
870 /* Store in the hsm_copy for later copytool use.
871 * Always modified even if no lsm. */
872 copy->hc_data_version = data_version;
876 /* On error, the request should be considered as completed */
877 if (hpk.hpk_errval > 0)
878 hpk.hpk_flags |= HP_FLAG_COMPLETED;
880 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
883 /* Return first error */
884 RETURN(rc != 0 ? rc : rc2);
888 * Generic handler to do any post-copy work.
890 * It will send the last hsm_progress update to coordinator to inform it
891 * that copy is finished and whether it was successful or not.
894 * - for ARCHIVE request, it will sample the file data version and compare it
895 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
896 * will be considered as failed.
897 * - for RESTORE request, it will sample the file data version and send it to
898 * coordinator which is useful if the file was imported as 'released'.
900 * \return 0 on success.
902 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
904 struct ll_sb_info *sbi = ll_s2sbi(sb);
905 struct hsm_progress_kernel hpk;
910 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
911 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
912 * initialized if copy_end was called with copy == NULL.
915 /* Forge a hsm_progress based on data from copy. */
916 hpk.hpk_fid = copy->hc_hai.hai_fid;
917 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
918 hpk.hpk_extent = copy->hc_hai.hai_extent;
919 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
920 hpk.hpk_errval = copy->hc_errval;
921 hpk.hpk_data_version = 0;
923 /* For archive request, we need to check the file data was not changed.
925 * For restore request, we need to send the file data version, this is
926 * useful when the file was created using hsm_import.
928 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
929 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
930 (copy->hc_errval == 0)) {
932 __u64 data_version = 0;
934 /* Get lsm for this fid */
935 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
937 hpk.hpk_flags |= HP_FLAG_RETRY;
938 /* hpk_errval must be >= 0 */
939 hpk.hpk_errval = -PTR_ERR(inode);
940 GOTO(progress, rc = PTR_ERR(inode));
943 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
946 CDEBUG(D_HSM, "Could not read file data version. "
947 "Request could not be confirmed.\n");
948 if (hpk.hpk_errval == 0)
949 hpk.hpk_errval = -rc;
953 /* Store in the hsm_copy for later copytool use.
954 * Always modified even if no lsm. */
955 hpk.hpk_data_version = data_version;
957 /* File could have been stripped during archiving, so we need
958 * to check anyway. */
959 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
960 (copy->hc_data_version != data_version)) {
961 CDEBUG(D_HSM, "File data version mismatched. "
962 "File content was changed during archiving. "
963 DFID", start:%#llx current:%#llx\n",
964 PFID(©->hc_hai.hai_fid),
965 copy->hc_data_version, data_version);
966 /* File was changed, send error to cdt. Do not ask for
967 * retry because if a file is modified frequently,
968 * the cdt will loop on retried archive requests.
969 * The policy engine will ask for a new archive later
970 * when the file will not be modified for some tunable
972 hpk.hpk_flags &= ~HP_FLAG_RETRY;
974 /* hpk_errval must be >= 0 */
975 hpk.hpk_errval = -rc;
982 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
985 /* Return first error */
986 RETURN(rc != 0 ? rc : rc2);
990 static int copy_and_ioctl(int cmd, struct obd_export *exp,
991 const void __user *data, size_t size)
996 OBD_ALLOC(copy, size);
1000 if (copy_from_user(copy, data, size)) {
1005 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1007 OBD_FREE(copy, size);
1012 static int check_owner(int type, int id)
1016 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1020 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1029 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1031 int cmd = qctl->qc_cmd;
1032 int type = qctl->qc_type;
1033 int id = qctl->qc_id;
1034 int valid = qctl->qc_valid;
1041 case LUSTRE_Q_SETDEFAULT:
1042 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1046 case LUSTRE_Q_GETDEFAULT:
1047 if (check_owner(type, id) &&
1048 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1054 CERROR("unsupported quotactl op: %#x\n", cmd);
1058 if (valid != QC_GENERAL) {
1059 if (cmd == Q_GETINFO)
1060 qctl->qc_cmd = Q_GETOINFO;
1061 else if (cmd == Q_GETQUOTA)
1062 qctl->qc_cmd = Q_GETOQUOTA;
1068 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1069 sizeof(*qctl), qctl, NULL);
1072 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1073 sizeof(*qctl), qctl, NULL);
1076 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1077 sizeof(*qctl), qctl, NULL);
1079 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1081 sizeof(*qctl), qctl, NULL);
1093 struct obd_quotactl *oqctl;
1095 OBD_ALLOC_PTR(oqctl);
1099 QCTL_COPY(oqctl, qctl);
1100 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1102 OBD_FREE_PTR(oqctl);
1105 /* If QIF_SPACE is not set, client should collect the
1106 * space usage from OSSs by itself */
1107 if (cmd == Q_GETQUOTA &&
1108 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1109 !oqctl->qc_dqblk.dqb_curspace) {
1110 struct obd_quotactl *oqctl_tmp;
1112 OBD_ALLOC_PTR(oqctl_tmp);
1113 if (oqctl_tmp == NULL)
1114 GOTO(out, rc = -ENOMEM);
1116 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1117 oqctl_tmp->qc_id = oqctl->qc_id;
1118 oqctl_tmp->qc_type = oqctl->qc_type;
1120 /* collect space usage from OSTs */
1121 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1122 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1123 if (!rc || rc == -EREMOTEIO) {
1124 oqctl->qc_dqblk.dqb_curspace =
1125 oqctl_tmp->qc_dqblk.dqb_curspace;
1126 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1129 /* collect space & inode usage from MDTs */
1130 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1131 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1132 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1133 if (!rc || rc == -EREMOTEIO) {
1134 oqctl->qc_dqblk.dqb_curspace +=
1135 oqctl_tmp->qc_dqblk.dqb_curspace;
1136 oqctl->qc_dqblk.dqb_curinodes =
1137 oqctl_tmp->qc_dqblk.dqb_curinodes;
1138 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1140 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1143 OBD_FREE_PTR(oqctl_tmp);
1146 QCTL_COPY(qctl, oqctl);
1147 OBD_FREE_PTR(oqctl);
1153 /* This function tries to get a single name component,
1154 * to send to the server. No actual path traversal involved,
1155 * so we limit to NAME_MAX */
1156 static char *ll_getname(const char __user *filename)
1161 OBD_ALLOC(tmp, NAME_MAX + 1);
1164 return ERR_PTR(-ENOMEM);
1166 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1169 else if (len > NAME_MAX)
1170 ret = -ENAMETOOLONG;
1173 OBD_FREE(tmp, NAME_MAX + 1);
1179 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1181 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1183 struct dentry *dentry = file_dentry(file);
1184 struct inode *inode = file_inode(file);
1185 struct ll_sb_info *sbi = ll_i2sbi(inode);
1186 struct obd_ioctl_data *data;
1190 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1191 PFID(ll_inode2fid(inode)), inode, cmd);
1193 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1194 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1197 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1199 case FS_IOC_GETFLAGS:
1200 case FS_IOC_SETFLAGS:
1201 RETURN(ll_iocontrol(inode, file, cmd, arg));
1202 case FSFILT_IOC_GETVERSION:
1203 case FS_IOC_GETVERSION:
1204 RETURN(put_user(inode->i_generation, (int __user *)arg));
1205 /* We need to special case any other ioctls we want to handle,
1206 * to send them to the MDS/OST as appropriate and to properly
1207 * network encode the arg field. */
1208 case FS_IOC_SETVERSION:
1211 case LL_IOC_GET_MDTIDX: {
1214 mdtidx = ll_get_mdt_idx(inode);
1218 if (put_user((int)mdtidx, (int __user *)arg))
1223 case IOC_MDC_LOOKUP: {
1224 int namelen, len = 0;
1228 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1233 filename = data->ioc_inlbuf1;
1234 namelen = strlen(filename);
1236 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1237 GOTO(out_free, rc = -EINVAL);
1240 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1242 CERROR("%s: lookup %.*s failed: rc = %d\n",
1243 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1248 OBD_FREE_LARGE(buf, len);
1251 case LL_IOC_LMV_SETSTRIPE: {
1252 struct lmv_user_md *lum;
1261 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1266 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1267 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1268 GOTO(lmv_out_free, rc = -EINVAL);
1270 filename = data->ioc_inlbuf1;
1271 namelen = data->ioc_inllen1;
1274 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1275 GOTO(lmv_out_free, rc = -EINVAL);
1277 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1278 lumlen = data->ioc_inllen2;
1280 if ((lum->lum_magic != LMV_USER_MAGIC &&
1281 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
1282 lumlen < sizeof(*lum)) {
1283 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1284 filename, lum->lum_magic, lumlen, -EFAULT);
1285 GOTO(lmv_out_free, rc = -EINVAL);
1288 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1289 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1291 mode = data->ioc_type;
1293 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1295 OBD_FREE_LARGE(buf, len);
1299 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1300 struct lmv_user_md lum;
1301 struct lmv_user_md __user *ulump =
1302 (struct lmv_user_md __user *)arg;
1305 if (copy_from_user(&lum, ulump, sizeof(lum)))
1308 if (lum.lum_magic != LMV_USER_MAGIC)
1311 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1315 case LL_IOC_LOV_SETSTRIPE_NEW:
1316 case LL_IOC_LOV_SETSTRIPE: {
1317 struct lov_user_md_v3 *lumv3 = NULL;
1318 struct lov_user_md_v1 lumv1;
1319 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1320 struct lov_user_md_v1 __user *lumv1p =
1321 (struct lov_user_md_v1 __user *)arg;
1322 struct lov_user_md_v3 __user *lumv3p =
1323 (struct lov_user_md_v3 __user *)arg;
1326 int set_default = 0;
1328 CLASSERT(sizeof(struct lov_user_md_v3) >
1329 sizeof(struct lov_comp_md_v1));
1330 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1331 /* first try with v1 which is smaller than v3 */
1332 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1335 if (inode->i_sb->s_root == file_dentry(file))
1338 switch (lumv1.lmm_magic) {
1339 case LOV_USER_MAGIC_V3:
1340 case LOV_USER_MAGIC_SPECIFIC:
1341 lum_size = ll_lov_user_md_size(&lumv1);
1344 OBD_ALLOC(lumv3, lum_size);
1347 if (copy_from_user(lumv3, lumv3p, lum_size))
1348 GOTO(out, rc = -EFAULT);
1349 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1351 case LOV_USER_MAGIC_V1:
1354 GOTO(out, rc = -ENOTSUPP);
1357 /* in v1 and v3 cases lumv1 points to data */
1358 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1361 OBD_FREE(lumv3, lum_size);
1364 case LL_IOC_LMV_GETSTRIPE: {
1365 struct lmv_user_md __user *ulmv =
1366 (struct lmv_user_md __user *)arg;
1367 struct lmv_user_md lum;
1368 struct ptlrpc_request *request = NULL;
1369 union lmv_mds_md *lmm = NULL;
1372 struct lmv_user_md *tmp = NULL;
1376 int max_stripe_count;
1380 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1383 max_stripe_count = lum.lum_stripe_count;
1384 /* lum_magic will indicate which stripe the ioctl will like
1385 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1386 * is for default LMV stripe */
1387 if (lum.lum_magic == LMV_MAGIC_V1)
1388 valid |= OBD_MD_MEA;
1389 else if (lum.lum_magic == LMV_USER_MAGIC)
1390 valid |= OBD_MD_DEFAULT_MEA;
1394 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1397 GOTO(finish_req, rc);
1399 /* Get default LMV EA */
1400 if (lum.lum_magic == LMV_USER_MAGIC) {
1401 if (lmmsize > sizeof(*ulmv))
1402 GOTO(finish_req, rc = -EINVAL);
1404 if (copy_to_user(ulmv, lmm, lmmsize))
1405 GOTO(finish_req, rc = -EFAULT);
1407 GOTO(finish_req, rc);
1410 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1411 if (max_stripe_count < stripe_count) {
1412 lum.lum_stripe_count = stripe_count;
1413 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1414 GOTO(finish_req, rc = -EFAULT);
1415 GOTO(finish_req, rc = -E2BIG);
1418 lum_size = lmv_user_md_size(stripe_count,
1419 LMV_USER_MAGIC_SPECIFIC);
1420 OBD_ALLOC(tmp, lum_size);
1422 GOTO(finish_req, rc = -ENOMEM);
1424 mdt_index = ll_get_mdt_idx(inode);
1426 GOTO(out_tmp, rc = -ENOMEM);
1428 tmp->lum_magic = LMV_MAGIC_V1;
1429 tmp->lum_stripe_count = 0;
1430 tmp->lum_stripe_offset = mdt_index;
1431 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1432 for (i = 0; i < stripe_count; i++) {
1435 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1436 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1438 GOTO(out_tmp, rc = mdt_index);
1440 tmp->lum_objects[i].lum_mds = mdt_index;
1441 tmp->lum_objects[i].lum_fid = fid;
1442 tmp->lum_stripe_count++;
1445 if (copy_to_user(ulmv, tmp, lum_size))
1446 GOTO(out_tmp, rc = -EFAULT);
1448 OBD_FREE(tmp, lum_size);
1450 ptlrpc_req_finished(request);
1454 case LL_IOC_REMOVE_ENTRY: {
1455 char *filename = NULL;
1459 /* Here is a little hack to avoid sending REINT_RMENTRY to
1460 * unsupported server, which might crash the server(LU-2730),
1461 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1462 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1463 * server will support REINT_RMENTRY XXX*/
1464 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1467 filename = ll_getname((const char __user *)arg);
1468 if (IS_ERR(filename))
1469 RETURN(PTR_ERR(filename));
1471 namelen = strlen(filename);
1473 GOTO(out_rmdir, rc = -EINVAL);
1475 rc = ll_rmdir_entry(inode, filename, namelen);
1478 ll_putname(filename);
1481 case LL_IOC_LOV_SWAP_LAYOUTS:
1483 case IOC_OBD_STATFS:
1484 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1485 case LL_IOC_LOV_GETSTRIPE:
1486 case LL_IOC_LOV_GETSTRIPE_NEW:
1487 case LL_IOC_MDC_GETINFO:
1488 case IOC_MDC_GETFILEINFO:
1489 case IOC_MDC_GETFILESTRIPE: {
1490 struct ptlrpc_request *request = NULL;
1491 struct lov_user_md __user *lump;
1492 struct lov_mds_md *lmm = NULL;
1493 struct mdt_body *body;
1494 char *filename = NULL;
1497 if (cmd == IOC_MDC_GETFILEINFO ||
1498 cmd == IOC_MDC_GETFILESTRIPE) {
1499 filename = ll_getname((const char __user *)arg);
1500 if (IS_ERR(filename))
1501 RETURN(PTR_ERR(filename));
1503 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1504 &lmmsize, &request);
1506 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1511 body = req_capsule_server_get(&request->rq_pill,
1513 LASSERT(body != NULL);
1518 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1519 cmd == LL_IOC_MDC_GETINFO)) {
1527 if (cmd == IOC_MDC_GETFILESTRIPE ||
1528 cmd == LL_IOC_LOV_GETSTRIPE ||
1529 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1530 lump = (struct lov_user_md __user *)arg;
1532 struct lov_user_mds_data __user *lmdp;
1533 lmdp = (struct lov_user_mds_data __user *)arg;
1534 lump = &lmdp->lmd_lmm;
1538 /* If the file has no striping then zero out *lump so
1539 * that the caller isn't confused by garbage. */
1540 if (clear_user(lump, sizeof(*lump)))
1541 GOTO(out_req, rc = -EFAULT);
1542 } else if (copy_to_user(lump, lmm, lmmsize)) {
1543 if (copy_to_user(lump, lmm, sizeof(*lump)))
1544 GOTO(out_req, rc = -EFAULT);
1548 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1549 struct lov_user_mds_data __user *lmdp;
1552 st.st_dev = inode->i_sb->s_dev;
1553 st.st_mode = body->mbo_mode;
1554 st.st_nlink = body->mbo_nlink;
1555 st.st_uid = body->mbo_uid;
1556 st.st_gid = body->mbo_gid;
1557 st.st_rdev = body->mbo_rdev;
1558 st.st_size = body->mbo_size;
1559 st.st_blksize = PAGE_SIZE;
1560 st.st_blocks = body->mbo_blocks;
1561 st.st_atime = body->mbo_atime;
1562 st.st_mtime = body->mbo_mtime;
1563 st.st_ctime = body->mbo_ctime;
1564 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1568 lmdp = (struct lov_user_mds_data __user *)arg;
1569 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1570 GOTO(out_req, rc = -EFAULT);
1575 ptlrpc_req_finished(request);
1577 ll_putname(filename);
1580 case OBD_IOC_QUOTACTL: {
1581 struct if_quotactl *qctl;
1583 OBD_ALLOC_PTR(qctl);
1587 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1588 GOTO(out_quotactl, rc = -EFAULT);
1590 rc = quotactl_ioctl(sbi, qctl);
1593 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1600 case OBD_IOC_GETDTNAME:
1601 case OBD_IOC_GETMDNAME:
1602 RETURN(ll_get_obd_name(inode, cmd, arg));
1603 case LL_IOC_FLUSHCTX:
1604 RETURN(ll_flush_ctx(inode));
1605 case LL_IOC_GETOBDCOUNT: {
1607 struct obd_export *exp;
1609 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1612 /* get ost count when count is zero, get mdt count otherwise */
1613 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1614 vallen = sizeof(count);
1615 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1616 KEY_TGT_COUNT, &vallen, &count);
1618 CERROR("get target count failed: %d\n", rc);
1622 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1627 case LL_IOC_PATH2FID:
1628 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1629 sizeof(struct lu_fid)))
1632 case LL_IOC_GET_CONNECT_FLAGS: {
1633 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1634 (void __user *)arg));
1636 case OBD_IOC_FID2PATH:
1637 RETURN(ll_fid2path(inode, (void __user *)arg));
1638 case LL_IOC_GETPARENT:
1639 RETURN(ll_getparent(file, (void __user *)arg));
1640 case LL_IOC_FID2MDTIDX: {
1641 struct obd_export *exp = ll_i2mdexp(inode);
1645 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1649 /* Call mdc_iocontrol */
1650 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1651 (__u32 __user *)&index);
1657 case LL_IOC_HSM_REQUEST: {
1658 struct hsm_user_request *hur;
1665 /* We don't know the true size yet; copy the fixed-size part */
1666 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1671 /* Compute the whole struct size */
1672 totalsize = hur_len(hur);
1677 /* Final size will be more than double totalsize */
1678 if (totalsize >= MDS_MAXREQSIZE / 3)
1681 OBD_ALLOC_LARGE(hur, totalsize);
1685 /* Copy the whole struct */
1686 if (copy_from_user(hur, (void __user *)arg, totalsize))
1687 GOTO(out_hur, rc = -EFAULT);
1689 if (hur->hur_request.hr_action == HUA_RELEASE) {
1690 const struct lu_fid *fid;
1694 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1695 fid = &hur->hur_user_item[i].hui_fid;
1696 f = search_inode_for_lustre(inode->i_sb, fid);
1702 rc = ll_hsm_release(f);
1708 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1713 OBD_FREE_LARGE(hur, totalsize);
1717 case LL_IOC_HSM_PROGRESS: {
1718 struct hsm_progress_kernel hpk;
1719 struct hsm_progress hp;
1721 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1724 hpk.hpk_fid = hp.hp_fid;
1725 hpk.hpk_cookie = hp.hp_cookie;
1726 hpk.hpk_extent = hp.hp_extent;
1727 hpk.hpk_flags = hp.hp_flags;
1728 hpk.hpk_errval = hp.hp_errval;
1729 hpk.hpk_data_version = 0;
1731 /* File may not exist in Lustre; all progress
1732 * reported to Lustre root */
1733 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1737 case LL_IOC_HSM_CT_START:
1738 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1741 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1742 sizeof(struct lustre_kernelcomm));
1745 case LL_IOC_HSM_COPY_START: {
1746 struct hsm_copy *copy;
1749 OBD_ALLOC_PTR(copy);
1752 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1757 rc = ll_ioc_copy_start(inode->i_sb, copy);
1758 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1764 case LL_IOC_HSM_COPY_END: {
1765 struct hsm_copy *copy;
1768 OBD_ALLOC_PTR(copy);
1771 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1776 rc = ll_ioc_copy_end(inode->i_sb, copy);
1777 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1783 case LL_IOC_MIGRATE: {
1784 struct lmv_user_md *lum;
1791 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1795 data = (struct obd_ioctl_data *)buf;
1796 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1797 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1798 GOTO(migrate_free, rc = -EINVAL);
1800 filename = data->ioc_inlbuf1;
1801 namelen = data->ioc_inllen1;
1803 if (namelen < 1 || namelen != strlen(filename) + 1) {
1804 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1805 GOTO(migrate_free, rc = -EINVAL);
1808 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1809 if (lum->lum_magic != LMV_USER_MAGIC &&
1810 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1812 CERROR("%s: wrong lum magic %x: rc = %d\n",
1813 filename, lum->lum_magic, rc);
1814 GOTO(migrate_free, rc);
1817 rc = ll_migrate(inode, file, lum, filename);
1819 OBD_FREE_LARGE(buf, len);
1823 case LL_IOC_FSGETXATTR:
1824 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1825 case LL_IOC_FSSETXATTR:
1826 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1828 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1829 (void __user *)arg));
1833 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1835 struct inode *inode = file->f_mapping->host;
1836 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1837 struct ll_sb_info *sbi = ll_i2sbi(inode);
1838 int api32 = ll_need_32bit_api(sbi);
1839 loff_t ret = -EINVAL;
1847 offset += file->f_pos;
1853 offset += LL_DIR_END_OFF_32BIT;
1855 offset += LL_DIR_END_OFF;
1862 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1863 (!api32 && offset <= LL_DIR_END_OFF))) {
1864 if (offset != file->f_pos) {
1865 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1866 (!api32 && offset == LL_DIR_END_OFF))
1867 fd->lfd_pos = MDS_DIR_END_OFF;
1868 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1869 fd->lfd_pos = offset << 32;
1871 fd->lfd_pos = offset;
1872 file->f_pos = offset;
1873 file->f_version = 0;
1880 inode_unlock(inode);
1884 static int ll_dir_open(struct inode *inode, struct file *file)
1887 RETURN(ll_file_open(inode, file));
1890 static int ll_dir_release(struct inode *inode, struct file *file)
1893 RETURN(ll_file_release(inode, file));
1896 const struct file_operations ll_dir_operations = {
1897 .llseek = ll_dir_seek,
1898 .open = ll_dir_open,
1899 .release = ll_dir_release,
1900 .read = generic_read_dir,
1901 #ifdef HAVE_DIR_CONTEXT
1902 .iterate = ll_iterate,
1904 .readdir = ll_readdir,
1906 .unlocked_ioctl = ll_dir_ioctl,