Whamcloud - gitweb
2f5265016d570420fa1648f45dc5b5848d57416a
[fs/lustre-release.git] / lustre / utils / lfs.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  *
31  * lustre/utils/lfs.c
32  *
33  * Author: Peter J. Braam <braam@clusterfs.com>
34  * Author: Phil Schwan <phil@clusterfs.com>
35  * Author: Robert Read <rread@clusterfs.com>
36  */
37
38 /* for O_DIRECTORY */
39 #ifndef _GNU_SOURCE
40 #define _GNU_SOURCE
41 #endif
42
43 #include <stdlib.h>
44 #include <stdio.h>
45 #include <inttypes.h>
46 #include <getopt.h>
47 #include <string.h>
48 #include <mntent.h>
49 #include <unistd.h>
50 #include <errno.h>
51 #include <err.h>
52 #include <pwd.h>
53 #include <grp.h>
54 #include <sys/ioctl.h>
55 #include <sys/quota.h>
56 #include <sys/time.h>
57 #include <sys/types.h>
58 #include <sys/stat.h>
59 #include <sys/param.h>
60 #include <sys/xattr.h>
61 #include <fcntl.h>
62 #include <dirent.h>
63 #include <time.h>
64 #include <ctype.h>
65 #include <zlib.h>
66 #include <libgen.h>
67 #include <asm/byteorder.h>
68 #include "lfs_project.h"
69
70 #include <libcfs/util/string.h>
71 #include <libcfs/util/ioctl.h>
72 #include <libcfs/util/parser.h>
73 #include <libcfs/util/string.h>
74 #include <lustre/lustreapi.h>
75 #include <linux/lustre/lustre_ver.h>
76 #include <linux/lustre/lustre_param.h>
77 #include <linux/lnet/nidstr.h>
78 #include <lnetconfig/cyaml.h>
79 #include "lstddef.h"
80
81 /* all functions */
82 static int lfs_find(int argc, char **argv);
83 static int lfs_getstripe(int argc, char **argv);
84 static int lfs_getdirstripe(int argc, char **argv);
85 static int lfs_setdirstripe(int argc, char **argv);
86 static int lfs_rmentry(int argc, char **argv);
87 static int lfs_unlink_foreign(int argc, char **argv);
88 static int lfs_osts(int argc, char **argv);
89 static int lfs_mdts(int argc, char **argv);
90 static int lfs_df(int argc, char **argv);
91 static int lfs_getname(int argc, char **argv);
92 static int lfs_check(int argc, char **argv);
93 #ifdef HAVE_SYS_QUOTA_H
94 static int lfs_setquota(int argc, char **argv);
95 static int lfs_quota(int argc, char **argv);
96 static int lfs_project(int argc, char **argv);
97 #endif
98 static int lfs_flushctx(int argc, char **argv);
99 static int lfs_poollist(int argc, char **argv);
100 static int lfs_changelog(int argc, char **argv);
101 static int lfs_changelog_clear(int argc, char **argv);
102 static int lfs_fid2path(int argc, char **argv);
103 static int lfs_path2fid(int argc, char **argv);
104 static int lfs_rmfid(int argc, char **argv);
105 static int lfs_data_version(int argc, char **argv);
106 static int lfs_hsm_state(int argc, char **argv);
107 static int lfs_hsm_set(int argc, char **argv);
108 static int lfs_hsm_clear(int argc, char **argv);
109 static int lfs_hsm_action(int argc, char **argv);
110 static int lfs_hsm_archive(int argc, char **argv);
111 static int lfs_hsm_restore(int argc, char **argv);
112 static int lfs_hsm_release(int argc, char **argv);
113 static int lfs_hsm_remove(int argc, char **argv);
114 static int lfs_hsm_cancel(int argc, char **argv);
115 static int lfs_swap_layouts(int argc, char **argv);
116 static int lfs_mv(int argc, char **argv);
117 static int lfs_ladvise(int argc, char **argv);
118 static int lfs_getsom(int argc, char **argv);
119 static int lfs_heat_get(int argc, char **argv);
120 static int lfs_heat_set(int argc, char **argv);
121 static int lfs_mirror(int argc, char **argv);
122 static int lfs_mirror_list_commands(int argc, char **argv);
123 static int lfs_list_commands(int argc, char **argv);
124 static inline int lfs_mirror_resync(int argc, char **argv);
125 static inline int lfs_mirror_verify(int argc, char **argv);
126 static inline int lfs_mirror_read(int argc, char **argv);
127 static inline int lfs_mirror_write(int argc, char **argv);
128 static inline int lfs_mirror_copy(int argc, char **argv);
129 static int lfs_pcc_attach(int argc, char **argv);
130 static int lfs_pcc_attach_fid(int argc, char **argv);
131 static int lfs_pcc_detach(int argc, char **argv);
132 static int lfs_pcc_detach_fid(int argc, char **argv);
133 static int lfs_pcc_state(int argc, char **argv);
134 static int lfs_pcc(int argc, char **argv);
135 static int lfs_pcc_list_commands(int argc, char **argv);
136 static int lfs_migrate_to_dom(int fd, int fdv, char *name,
137                               __u64 migration_flags,
138                               struct llapi_stripe_param *param,
139                               struct llapi_layout *layout);
140
141 struct pool_to_id_cbdata {
142         const char *pool;
143         __u32 id;
144 };
145 static int find_comp_id_by_pool(struct llapi_layout *layout, void *cbdata);
146 static int find_mirror_id_by_pool(struct llapi_layout *layout, void *cbdata);
147
148 enum setstripe_origin {
149         SO_SETSTRIPE,
150         SO_MIGRATE,
151         SO_MIGRATE_MDT,
152         SO_MIRROR_CREATE,
153         SO_MIRROR_EXTEND,
154         SO_MIRROR_SPLIT,
155         SO_MIRROR_DELETE,
156 };
157
158 static int lfs_setstripe_internal(int argc, char **argv,
159                                   enum setstripe_origin opc);
160
161 static inline int lfs_setstripe(int argc, char **argv)
162 {
163         return lfs_setstripe_internal(argc, argv, SO_SETSTRIPE);
164 }
165
166 static inline int lfs_setstripe_migrate(int argc, char **argv)
167 {
168         return lfs_setstripe_internal(argc, argv, SO_MIGRATE);
169 }
170
171 static inline int lfs_mirror_create(int argc, char **argv)
172 {
173         return lfs_setstripe_internal(argc, argv, SO_MIRROR_CREATE);
174 }
175
176 static inline int lfs_mirror_extend(int argc, char **argv)
177 {
178         return lfs_setstripe_internal(argc, argv, SO_MIRROR_EXTEND);
179 }
180
181 static inline int lfs_mirror_split(int argc, char **argv)
182 {
183         return lfs_setstripe_internal(argc, argv, SO_MIRROR_SPLIT);
184 }
185
186 static inline int lfs_mirror_delete(int argc, char **argv)
187 {
188         return lfs_setstripe_internal(argc, argv, SO_MIRROR_DELETE);
189 }
190
191 /* Setstripe and migrate share mostly the same parameters */
192 #define SSM_CMD_COMMON(cmd) \
193         "usage: "cmd" [--component-end|-E COMP_END]\n"                  \
194         "                 [--stripe-count|-c STRIPE_COUNT]\n"           \
195         "                 [--overstripe-count|-C STRIPE_COUNT]\n"       \
196         "                 [--stripe-index|-i START_OST_IDX]\n"          \
197         "                 [--stripe-size|-S STRIPE_SIZE]\n"             \
198         "                 [--extension-size|--ext-size|-z]\n"           \
199         "                 [--help|-h] [--layout|-L PATTERN]\n"          \
200         "                 [--mirror_count|-N[MIRROR_COUNT]]\n"          \
201         "                 [--ost|-o OST_INDICES]\n"                     \
202         "                 [--pool|-p POOL_NAME]\n"                      \
203         "                 [--yaml|-y YAML_TEMPLATE_FILE]\n"             \
204         "                 [--copy=LUSTRE_SRC]\n"
205
206 #define SSM_HELP_COMMON \
207         "\tSTRIPE_COUNT: Number of OSTs to stripe on (0=fs default, -1 all)\n" \
208         "\t              Using -C instead of -c allows overstriping, which\n"  \
209         "\t              will place more than one stripe per OST if\n"         \
210         "\t              stripe_count is greater than the number of OSTs.\n"   \
211         "\tSTART_OST_IDX: OST index of first stripe (-1=default round robin)\n"\
212         "\tSTRIPE_SIZE:  Number of bytes on each OST (0=fs default)\n"         \
213         "\t              Optional K, M, or G suffix (for KB, MB, GB\n"         \
214         "\t              respectively).  Must be a multiple of 64KiB.\n"       \
215         "\tEXTENSION_SIZE:\n"                                                  \
216         "\t              Number of bytes the previous component is extended\n" \
217         "\t              each time. Optional K, M, or G suffix (for KB,\n"     \
218         "\t              MB, GB respectively)\n"                               \
219         "\tPOOL_NAME:    Name of OST pool to use (default none)\n"             \
220         "\tLAYOUT:       stripe pattern type: raid0, mdt (default raid0)\n"    \
221         "\tOST_INDICES:  List of OST indices, can be repeated multiple times\n"\
222         "\t              Indices be specified in a format of:\n"               \
223         "\t                -o OST_1,OST_I-OST_J,OST_N\n"                       \
224         "\t              Or:\n"                                                \
225         "\t                -o OST_1 -o OST_I-OST_J -o OST_N\n"                 \
226         "\t              If --pool is set with --ost then the OSTs\n"          \
227         "\t              must be the members of the pool.\n"                   \
228         "\tCOMP_END:     Extent end of component, start after previous end.\n" \
229         "\t              Optional K, M, or G suffix (for KiB, MiB, GiB), or\n" \
230         "\t              -1 or 'eof' for max file size). Must be a multiple\n" \
231         "\t              of stripe_size and a multiple of 64KiB.\n"            \
232         "\tYAML_TEMPLATE_FILE:\n"                                              \
233         "\t              YAML layout template file, can't be used with -c,\n"  \
234         "\t              -i, -S, -p, -o, or -E arguments.\n"                   \
235         "\tLUSTRE_SRC:   Lustre file/dir whose layout info is used to set\n"   \
236         "\t              another lustre file or directory, can't used with\n"  \
237         "\t              -c, -i, -S, -p, -o, or -E arguments.\n"
238
239 #define MIRROR_CREATE_HELP                                                     \
240         "\tMIRROR_COUNT: Number of mirrors to be created with the upcoming\n"  \
241         "\t              setstripe layout options\n"                           \
242         "\t              It defaults to 1 if not specified; if specified,\n"   \
243         "\t              it must follow the option without a space.\n"         \
244         "\t              The option can also be repeated multiple times to\n"  \
245         "\t              separate mirrors that have different layouts.\n"      \
246         "\tSETSTRIPE_OPTIONS: Mirror layout as with 'setstripe'\n"             \
247         "\t              It can be a plain layout or a composite layout.\n"    \
248         "\t              If not specified, the stripe options inherited\n"     \
249         "\t              from the previous component will be used.\n"          \
250         "\tFLAGS:        set flags to the component of the current mirror.\n"  \
251         "\t              Only \"prefer\" flag is supported so far.\n"
252
253 #define MIRROR_EXTEND_HELP                                                     \
254         MIRROR_CREATE_HELP                                                     \
255         "\tVICTIM_FILE:  The layout of victim_file will be split and used\n"   \
256         "\t              as a mirror added to the mirrored file.\n"            \
257         "\tno-verify:    This option indicates not to verify the mirror(s)\n"  \
258         "\t              from victim file(s) in case the victim file(s)\n"     \
259         "\t              contains the same data as the original mirrored\n"    \
260         "\t              file.\n"
261
262 #define MIRROR_EXTEND_USAGE                                                    \
263         "                 {--mirror-count|-N[mirror_count]}\n"                 \
264         "                 [SETSTRIPE_OPTIONS|-f|--file VICTIM_FILE]\n"         \
265         "                 [--no-verify]\n"
266
267 #define SETSTRIPE_USAGE                                                 \
268         SSM_CMD_COMMON("setstripe")                                     \
269         MIRROR_EXTEND_USAGE                                             \
270         "                 <directory|filename>\n"                       \
271         SSM_HELP_COMMON                                                 \
272         MIRROR_EXTEND_HELP
273
274 #define MIGRATE_USAGE                                                   \
275         SSM_CMD_COMMON("migrate  ")                                     \
276         "                 [--block|-b] [--non-block|-n]\n"              \
277         "                 [--non-direct|-D] [--verbose|-v]\n"           \
278         "                 <filename>\n"                                 \
279         SSM_HELP_COMMON                                                 \
280         "\n"                                                            \
281         "\tblock:        Block file access during data migration (default)\n" \
282         "\tnon-block:    Abort migrations if concurrent access is detected\n" \
283         "\tnon-direct:   Do not use direct I/O to copy file contents\n" \
284         "\tverbose:      Print each filename as it is migrated\n"       \
285
286 #define SETDIRSTRIPE_USAGE                                              \
287         "               [--mdt-count|-c stripe_count>\n"                \
288         "               [--help|-h] [--mdt-hash|-H mdt_hash]\n"         \
289         "               [--mdt-index|-i mdt_index[,mdt_index,...]\n"    \
290         "               [--default|-D] [--mode|-o mode]\n"              \
291         "               [--max-inherit|-X max_inherit]\n"               \
292         "               [--max-inherit-rr max_inherit_rr] <dir>\n"      \
293         "\tstripe_count: stripe count of the striped directory\n"       \
294         "\tmdt_index: MDT index of first stripe\n"                      \
295         "\tmdt_hash:  hash type of the striped directory. mdt types:\n" \
296         "       crush     CRUSH hash algorithm (default)\n" \
297         "       fnv_1a_64 FNV-1a hash algorithm\n"              \
298         "       all_char  sum of characters % MDT_COUNT (not recommended)\n" \
299         "\tdefault_stripe: set default dirstripe of the directory\n"    \
300         "\tmode: the file access permission of the directory (octal)\n" \
301         "To create dir with a foreign (free format) layout :\n"         \
302         "setdirstripe|mkdir --foreign[=FOREIGN_TYPE] -x|-xattr STRING " \
303                 "[--mode|-o MODE] [--flags HEX] <dir>\n" \
304         "\tmode: the file access permissions of the directory\n"        \
305         "\tforeign_type: none, daos, symlink, or numeric value\n"
306
307 /**
308  * command_t mirror_cmdlist - lfs mirror commands.
309  */
310 command_t mirror_cmdlist[] = {
311         { .pc_name = "create", .pc_func = lfs_mirror_create,
312           .pc_help = "Create a mirrored file.\n"
313                 "usage: lfs mirror create {--mirror-count|-N[MIRROR_COUNT]}\n"
314                 "\t\t[SETSTRIPE_OPTIONS] <filename|directory> ...\n"
315           MIRROR_CREATE_HELP },
316         { .pc_name = "delete", .pc_func = lfs_mirror_delete,
317           .pc_help = "Delete a mirror from a file.\n"
318         "usage: lfs mirror delete {--mirror-id MIRROR_ID|-p POOL|\n"
319         "\t               --component-id|--comp-id|-I COMP_ID} <mirror_file>\n"
320         },
321         { .pc_name = "extend", .pc_func = lfs_mirror_extend,
322           .pc_help = "Extend a mirrored file.\n"
323                 "usage: lfs mirror extend {--mirror-count|-N[MIRROR_COUNT]}\n"
324                 "\t\t[--no-verify] [SETSTRIPE_OPTIONS|-f VICTIM_FILE] ... <filename> ...\n"
325           MIRROR_EXTEND_HELP },
326         { .pc_name = "split", .pc_func = lfs_mirror_split,
327           .pc_help = "Split a mirrored file.\n"
328         "usage: lfs mirror split [--destroy|-d] [-f NEW_FILE] [--help|-h]\n"
329         "\t\t{--mirror-id MIRROR_ID|--component-id|-I COMP_ID|-p POOL}\n"
330         "\t\t<mirrored file> ...\n"
331         "\tMIRROR_ID: The numerical unique identifier for a mirror.\n"
332         "\t           It can be fetched by the 'lfs getstripe' command.\n"
333         "\tCOMP_ID:   Unique component ID within a mirror.\n"
334         "\tPOOL:      Components using specified pool.\n"
335         "\tNEW_FILE:  This option indicates the layout of the split\n"
336         "\t           mirror will be stored into. If not specified,\n"
337         "\t           a new file named <mirrored_file>.mirror~MIRROR_ID\n"
338         "\t           will be used.\n" },
339         { .pc_name = "read", .pc_func = lfs_mirror_read,
340           .pc_help = "Read the content of a specified mirror of a file.\n"
341                 "usage: lfs mirror read {--mirror-id|-N MIRROR_ID}\n"
342                 "\t\t[--outfile|-o <output_file>] <mirrored_file>\n" },
343         { .pc_name = "write", .pc_func = lfs_mirror_write,
344           .pc_help = "Write to a specified mirror of a file.\n"
345                 "usage: lfs mirror write {--mirror-id|-N MIRROR_ID}\n"
346                 "\t\t[--inputfile|-i <input_file>] <mirrored_file>\n" },
347         { .pc_name = "copy", .pc_func = lfs_mirror_copy,
348           .pc_help = "Copy a specified mirror to other mirror(s) of a file.\n"
349                 "usage: lfs mirror copy {--read-mirror|-i MIRROR_ID0}\n"
350                 "\t\t{--write-mirror|-o MIRROR_ID1[,...]} <mirrored_file>\n" },
351         { .pc_name = "resync", .pc_func = lfs_mirror_resync,
352           .pc_help = "Resynchronizes out-of-sync mirrored file(s).\n"
353                 "usage: lfs mirror resync [--only MIRROR_ID[,...]>]\n"
354                 "\t\t<mirrored_file> [<mirrored_file2>...]\n" },
355         { .pc_name = "verify", .pc_func = lfs_mirror_verify,
356           .pc_help = "Verify mirrored file(s).\n"
357                 "usage: lfs mirror verify [--only MIRROR_ID[,...]]\n"
358                 "\t\t[--verbose|-v] <mirrored_file> [<mirrored_file2> ...]\n" },
359         { .pc_name = "list-commands", .pc_func = lfs_mirror_list_commands,
360           .pc_help = "list commands supported by lfs mirror"},
361         { .pc_name = "help", .pc_func = Parser_help, .pc_help = "help" },
362         { .pc_name = "exit", .pc_func = Parser_quit, .pc_help = "quit" },
363         { .pc_name = "quit", .pc_func = Parser_quit, .pc_help = "quit" },
364         { .pc_help = NULL }
365 };
366
367 /**
368  * command_t pcc_cmdlist - lfs pcc commands.
369  */
370 command_t pcc_cmdlist[] = {
371         { .pc_name = "attach", .pc_func = lfs_pcc_attach,
372           .pc_help = "Attach given files to the Persistent Client Cache.\n"
373                 "usage: lfs pcc attach <--id|-i NUM> <file> ...\n"
374                 "\t-i: archive id for RW-PCC\n" },
375         { .pc_name = "attach_fid", .pc_func = lfs_pcc_attach_fid,
376           .pc_help = "Attach given files into PCC by FID(s).\n"
377                 "usage: lfs pcc attach_id {--id|-i NUM} {--mnt|-m MOUNTPOINT} FID ...\n"
378                 "\t-i: archive id for RW-PCC\n"
379                 "\t-m: Lustre mount point\n" },
380         { .pc_name = "state", .pc_func = lfs_pcc_state,
381           .pc_help = "Display the PCC state for given files.\n"
382                 "usage: lfs pcc state <file> ...\n" },
383         { .pc_name = "detach", .pc_func = lfs_pcc_detach,
384           .pc_help = "Detach given files from the Persistent Client Cache.\n"
385                 "usage: lfs pcc detach <file> ...\n" },
386         { .pc_name = "detach_fid", .pc_func = lfs_pcc_detach_fid,
387           .pc_help = "Detach given files from PCC by FID(s).\n"
388                 "usage: lfs pcc detach_fid <mntpath> <fid>...\n" },
389         { .pc_name = "list-commands", .pc_func = lfs_pcc_list_commands,
390           .pc_help = "list commands supported by lfs pcc"},
391         { .pc_name = "help", .pc_func = Parser_help, .pc_help = "help" },
392         { .pc_name = "exit", .pc_func = Parser_quit, .pc_help = "quit" },
393         { .pc_name = "quit", .pc_func = Parser_quit, .pc_help = "quit" },
394         { .pc_help = NULL }
395 };
396
397 /* all available commands */
398 command_t cmdlist[] = {
399         {"setstripe", lfs_setstripe, 0,
400          "To create a file with specified striping/composite layout, or\n"
401          "create/replace the default layout on an existing directory:\n"
402          SSM_CMD_COMMON("setstripe")
403          "                 [--mode MODE]\n"
404          "                 <directory|filename>\n"
405          " or\n"
406          "To add component(s) to an existing composite file:\n"
407          SSM_CMD_COMMON("setstripe --component-add")
408          SSM_HELP_COMMON
409          "To totally delete the default striping from an existing directory:\n"
410          "usage: setstripe [--delete|-d] <directory>\n"
411          " or\n"
412          "To create a mirrored file or set s default mirror layout on a directory:\n"
413          "usage: setstripe {--mirror-count|-N}[MIRROR_COUNT] [SETSTRIPE_OPTIONS] <directory|filename>\n"
414          " or\n"
415          "To delete the last component(s) from an existing composite file\n"
416          "(note that this will also delete any data in those components):\n"
417          "usage: setstripe --component-del [--component-id|-I COMP_ID]\n"
418          "                               [--component-flags|-F COMP_FLAGS]\n"
419          "                               <filename>\n"
420          "\tCOMP_ID:     Unique component ID to delete\n"
421          "\tCOMP_FLAGS:  'init' indicating all instantiated components\n"
422          "\t             '^init' indicating all uninstantiated components\n"
423          "\t-I and -F cannot be specified at the same time\n"
424          " or\n"
425          "To set or clear flags on a specific component\n"
426          "(note that this command can only be applied to mirrored files:\n"
427          "usage: setstripe --comp-set {-I COMP_ID|--comp-flags=COMP_FLAGS}\n"
428          "                            <filename>\n"
429          " or\n"
430          "To create a file with a foreign (free format) layout:\n"
431          "usage: setstripe --foreign[=FOREIGN_TYPE]\n"
432          "                 --xattr|-x LAYOUT_STRING [--flags HEX]\n"
433          "                 [--mode MODE] <filename>\n"},
434         {"getstripe", lfs_getstripe, 0,
435          "To list the layout pattern for a given file or files in a\n"
436          "directory or recursively for all files in a directory tree.\n"
437          "usage: getstripe [--ost|-O UUID] [--quiet|-q] [--verbose|-v]\n"
438          "                 [--stripe-count|-c] [--stripe-index|-i] [--fid|-F]\n"
439          "                 [--pool|-p] [--stripe-size|-S] [--directory|-d]\n"
440          "                 [--mdt-index|-m] [--recursive|-r] [--raw|-R]\n"
441          "                 [--layout|-L] [--generation|-g] [--yaml|-y]\n"
442          "                 [--help|-h] [--component-id|-I[=COMP_ID]]\n"
443          "                 [--component-flags[=COMP_FLAGS]]\n"
444          "                 [--component-count]\n"
445          "                 [--extension-size|--ext-size|-z]\n"
446          "                 [--component-start[=[+-]COMP_START]]\n"
447          "                 [--component-end[=[+-]COMP_END]|-E[[+-]comp_end]]\n"
448          "                 [[!] --mirror-index=[+-]INDEX |\n"
449          "                 [!] --mirror-id=[+-]MIRROR_ID] [--mirror-count|-N]\n"
450          "                 <directory|filename> ..."},
451         {"setdirstripe", lfs_setdirstripe, 0,
452          "Create striped directory on specified MDT, same as mkdir.\n"
453          "May be restricted to root or group users, depending on settings.\n"
454          "usage: setdirstripe [OPTION] <directory>\n"
455          SETDIRSTRIPE_USAGE},
456         {"getdirstripe", lfs_getdirstripe, 0,
457          "To list the layout pattern info for a given directory\n"
458          "or recursively for all directories in a directory tree.\n"
459          "usage: getdirstripe [--mdt-count|-c] [--mdt-index|-m|-i]\n"
460          "                    [--help|-h] [--mdt-hash|-H] [--obd|-O UUID]\n"
461          "                    [--recursive|-r] [--yaml|-y]\n"
462          "                    [--verbose|-v] [--default|-D]\n"
463          "                    [--max-inherit|-X]\n"
464          "                    [--max-inherit-rr] <dir> ..."},
465         {"mkdir", lfs_setdirstripe, 0,
466          "Create striped directory on specified MDT, same as setdirstripe.\n"
467          "usage: mkdir [OPTION] <directory>\n"
468          SETDIRSTRIPE_USAGE},
469         {"rm_entry", lfs_rmentry, 0,
470          "To remove the name entry of the remote directory. Note: This\n"
471          "command will only delete the name entry, i.e. the remote directory\n"
472          "will become inaccessable after this command. This can only be done\n"
473          "by the administrator\n"
474          "usage: rm_entry <dir>\n"},
475         {"unlink_foreign", lfs_unlink_foreign, 0,
476          "To remove the foreign file/dir.\n"
477          "Note: This is for files/dirs prevented to be removed using\n"
478          "unlink/rmdir, but works also for regular ones\n"
479          "usage: unlink_foreign <foreign_dir/file> [<foreign_dir/file> ...]\n"},
480         {"pool_list", lfs_poollist, 0,
481          "List pools or pool OSTs\n"
482          "usage: pool_list <fsname>[.<pool>] | <pathname>\n"},
483         {"find", lfs_find, 0,
484          "find files matching given attributes recursively in directory tree.\n"
485          "usage: find <directory|filename> ...\n"
486          "     [[!] --atime|-A [+-]N[smhdwy]] [[!] --ctime|-C [+-]N[smhdwy]]\n"
487          "     [[!] --mtime|-M [+-]N[smhdwy]]\n"
488          "     [[!] --btime|--Btime|-B [+-]N[smhdwy]] [--help|-h]\n"
489          "     [[!] --newer[XY] <reference>] [[!] --blocks|-b N]\n"
490          "     [--maxdepth|-D N] [[!] --mdt-index|--mdt|-m <uuid|index,...>]\n"
491          "     [[!] --name|-n <pattern>] [[!] --ost|-O <uuid|index,...>]\n"
492          "     [[!] --perm [/-]mode] [[!] --pool <pool>] [--print|-P]\n"
493          "     [--print0|-0] [[!] --projid <projid>]\n"
494          "     [[!] --size|-s [+-]N[bkMGTPE]]\n"
495          "     [[!] --stripe-count|-c [+-]<stripes>]\n"
496          "     [[!] --stripe-index|-i <index,...>]\n"
497          "     [[!] --stripe-size|-S [+-]N[kMGT]] [[!] --type|-t <filetype>]\n"
498          "     [[!] --extension-size|--ext-size|-z [+-]N[kMGT]]\n"
499          "     [[!] --gid|-g|--group|-G <gid>|<gname>]\n"
500          "     [[!] --uid|-u|--user|-U <uid>|<uname>]\n"
501          "     [[!] --layout|-L released,raid0,mdt]\n"
502          "     [[!] --foreign[=<foreign_type>]]\n"
503          "     [[!] --component-count [+-]<comp_cnt>]\n"
504          "     [[!] --component-start [+-]N[kMGTPE]]\n"
505          "     [[!] --component-end|-E [+-]N[kMGTPE]]\n"
506          "     [[!] --component-flags {init,stale,prefer,offline,nosync,extension}]\n"
507          "     [[!] --mirror-count|-N [+-]<n>]\n"
508          "     [[!] --mirror-state <[^]state>]\n"
509          "     [[!] --mdt-count|-T [+-]<stripes>]\n"
510          "     [[!] --mdt-hash|-H <[^][blm],[^]fnv_1a_64,all_char,crush,...>\n"
511          "     [[!] --mdt-index|-m <uuid|index,...>]\n"
512          "\t !: used before an option indicates 'NOT' requested attribute\n"
513          "\t -: used before a value indicates less than requested value\n"
514          "\t +: used before a value indicates more than requested value\n"
515          "\t ^: used before a flag indicates to exclude it\n"},
516         {"check", lfs_check, 0,
517          "Display the status of MGTs, MDTs or OSTs (as specified in the command)\n"
518          "or all the servers (MGTs, MDTs and OSTs).\n"
519          "usage: check {mgts|osts|mdts|all}"},
520         {"osts", lfs_osts, 0, "list OSTs connected to client "
521          "[for specified path only]\n" "usage: osts [path]"},
522         {"mdts", lfs_mdts, 0, "list MDTs connected to client "
523          "[for specified path only]\n" "usage: mdts [path]"},
524         {"df", lfs_df, 0,
525          "report filesystem disk space usage or inodes usage "
526          "of each MDS and all OSDs or a batch belonging to a specific pool.\n"
527          "Usage: df [--inodes|-i] [--human-readable|-h] [--lazy|-l]\n"
528          "          [--pool|-p <fsname>[.<pool>]] [path]"},
529         {"getname", lfs_getname, 0,
530          "list instances and specified mount points [for specified path only]\n"
531          "Usage: getname [--help|-h] [--instance|-i] [--fsname|-n] [path ...]"},
532 #ifdef HAVE_SYS_QUOTA_H
533         {"setquota", lfs_setquota, 0, "Set filesystem quotas.\n"
534          "usage: setquota {-u|-g|-p} UNAME|UID|GNAME|GID|PROJID\n"
535          "                -b BLOCK_SOFTLIMIT -B BLOCK_HARDLIMIT\n"
536          "                -i INODE_SOFTLIMIT -I INODE_HARDLIMIT <filesystem>\n"
537          "       setquota {-u|--user USER|UID|-g|--group GROUP|GID|-p|--projid PROJID}\n"
538          "                [--block-softlimit BLOCK_SOFTLIMIT]\n"
539          "                [--block-hardlimit BLOCK_HARDLIMIT]\n"
540          "                [--inode-softlimit INODE_SOFTLIMIT]\n"
541          "                [--inode-hardlimit INODE_HARDLIMIT] <filesystem>\n"
542          "       setquota [-t] {-h|--help|-u|--user|-g|--group|-p|--projid}\n"
543          "                [--block-grace 'notify'|BLOCK_GRACE]\n"
544          "                [--inode-grace 'notify'|INODE_GRACE] <filesystem>\n"
545          "       setquota {-U|-G|-P}\n"
546          "                -b BLOCK_SOFTLIMIT -B BLOCK_HARDLIMIT\n"
547          "                -i INODE_SOFTLIMIT -I INODE_HARDLIMIT <filesystem>\n"
548          "       setquota {-U|--default-usr|-G|--default-grp|-P|--default-prj}\n"
549          "                [--block-softlimit BLOCK_SOFTLIMIT]\n"
550          "                [--block-hardlimit BLOCK_HARDLIMIT]\n"
551          "                [--inode-softlimit INODE_SOFTLIMIT]\n"
552          "                [--inode-hardlimit INODE_HARDLIMIT] <filesystem>\n"
553          "       setquota {-u|-g|-p} UNAME|UID|GNAME|GID|PROJID\n"
554          "                {-d|--default}\n"
555          "       -b can be used instead of --block-softlimit/--block-grace\n"
556          "       -B can be used instead of --block-hardlimit\n"
557          "       -i can be used instead of --inode-softlimit/--inode-grace\n"
558          "       -I can be used instead of --inode-hardlimit\n"
559          "       -d can be used instead of --default\n\n"
560          "Note: The total quota space will be split into many qunits and\n"
561          "      balanced over all server targets, the minimal qunit size is\n"
562          "      1M bytes for block space and 1K inodes for inode space.\n\n"
563          "      The maximum quota grace time is 2^48 - 1 seconds.\n\n"
564          "      Quota space rebalancing process will stop when this mininum\n"
565          "      value is reached. As a result, quota exceeded can be returned\n"
566          "      while many targets still have 1MB or 1K inodes of spare\n"
567          "      quota space.\n\n"
568          "      When setting the grace time, 'notify' can be used as grace to\n"
569          "      be notified after the quota is over soft limit but prevents\n"
570          "      the soft limit from becoming the hard limit."},
571         {"quota", lfs_quota, 0, "Display disk usage and limits.\n"
572          "usage: quota [-q] [-v] [-h] [-o OBD_UUID|-i MDT_IDX|-I OST_IDX]\n"
573          "             [{-u|-g|-p} UNAME|UID|GNAME|GID|PROJID] <filesystem>\n"
574          "       quota [-o OBD_UUID|-i MDT_IDX|-I OST_IDX] -t {-u|-g|-p} <filesystem>\n"
575         "        quota [-q] [-v] [h] {-U|-G|-P} <filesystem>"},
576         {"project", lfs_project, 0,
577          "Change or list project attribute for specified file or directory.\n"
578          "usage: project [-d|-r] <file|directory...>\n"
579          "         list project ID and flags on file(s) or directories\n"
580          "       project [-p id] [-s] [-r] <file|directory...>\n"
581          "         set project ID and/or inherit flag for specified file(s) or directories\n"
582          "       project -c [-d|-r [-p id] [-0]] <file|directory...>\n"
583          "         check project ID and flags on file(s) or directories, print outliers\n"
584          "       project -C [-r] [-k] <file|directory...>\n"
585          "         clear the project inherit flag and ID on the file or directory\n"
586         },
587 #endif
588         {"flushctx", lfs_flushctx, 0,
589          "Flush security context for current user.\n"
590          "usage: flushctx [-k] [-r] [mountpoint...]"},
591         {"changelog", lfs_changelog, 0,
592          "Show the metadata changes on an MDT."
593          "\nusage: changelog <mdtname> [startrec [endrec]]"},
594         {"changelog_clear", lfs_changelog_clear, 0,
595          "Indicate that old changelog records up to <endrec> are no longer of "
596          "interest to consumer <id>, allowing the system to free up space.\n"
597          "An <endrec> of 0 means all records.\n"
598          "usage: changelog_clear <mdtname> <id> <endrec>"},
599         {"fid2path", lfs_fid2path, 0,
600          "Resolve the full path(s) for given FID(s). For a specific hardlink "
601          "specify link number <linkno>.\n"
602          "usage: fid2path [--print-fid|-f] [--print-link|-c] [--link|-l <linkno>] "
603          "<fsname|root> <fid>..."},
604         {"path2fid", lfs_path2fid, 0, "Display the fid(s) for a given path(s).\n"
605          "usage: path2fid [--parents] <path> ..."},
606         {"rmfid", lfs_rmfid, 0, "Remove file(s) by FID(s)\n"
607          "usage: rmfid <fsname|rootpath> <fid> ..."},
608         {"data_version", lfs_data_version, 0, "Display file data version for "
609          "a given path.\n" "usage: data_version [-n|-r|-w] <path>"},
610         {"hsm_state", lfs_hsm_state, 0, "Display the HSM information (states, "
611          "undergoing actions) for given files.\n usage: hsm_state <file> ..."},
612         {"hsm_set", lfs_hsm_set, 0, "Set HSM user flag on specified files.\n"
613          "usage: hsm_set [--norelease] [--noarchive] [--dirty] [--exists] "
614          "[--archived] [--lost] [--archive-id NUM] <file> ..."},
615         {"hsm_clear", lfs_hsm_clear, 0, "Clear HSM user flag on specified "
616          "files.\n"
617          "usage: hsm_clear [--norelease] [--noarchive] [--dirty] [--exists] "
618          "[--archived] [--lost] <file> ..."},
619         {"hsm_action", lfs_hsm_action, 0, "Display current HSM request for "
620          "given files.\n" "usage: hsm_action <file> ..."},
621         {"hsm_archive", lfs_hsm_archive, 0,
622          "Archive file to external storage.\n"
623          "usage: hsm_archive [--filelist FILELIST] [--data DATA] [--archive NUM] "
624          "<file> ..."},
625         {"hsm_restore", lfs_hsm_restore, 0,
626          "Restore file from external storage.\n"
627          "usage: hsm_restore [--filelist FILELIST] [--data DATA] <file> ..."},
628         {"hsm_release", lfs_hsm_release, 0,
629          "Release files from Lustre.\n"
630          "usage: hsm_release [--filelist FILELIST] [--data DATA] <file> ..."},
631         {"hsm_remove", lfs_hsm_remove, 0,
632          "Remove file copy from external storage.\n"
633          "usage: hsm_remove [--filelist FILELIST] [--data DATA] "
634          "[--archive NUM]\n"
635          "                  (FILE [FILE ...] | "
636          "--mntpath MOUNTPATH FID [FID ...])\n"
637          "\n"
638          "Note: To remove an archived copy of a file already deleted from a "
639          "Lustre FS, the\n"
640          "--mntpath option and a list of FIDs must be specified"
641         },
642         {"hsm_cancel", lfs_hsm_cancel, 0,
643          "Cancel requests related to specified files.\n"
644          "usage: hsm_cancel [--filelist FILELIST] [--data DATA] <file> ..."},
645         {"swap_layouts", lfs_swap_layouts, 0, "Swap layouts between 2 files.\n"
646          "usage: swap_layouts <path1> <path2>"},
647         {"migrate", lfs_setstripe_migrate, 0,
648          "migrate a directory between MDTs.\n"
649          "usage: migrate [--mdt-count|-c STRIPE_COUNT]\n"
650          "               [--mdt-hash|-H HASH_TYPE]\n"
651          "               [--mdt-index|-m START_MDT_INDEX] [--verbose|-v]\n"
652          "               <directory>\n"
653          "\tmdt:        MDTs to stripe over, if only one MDT is specified\n"
654          "                      it's the MDT index of first stripe\n"
655          "\tmdt_count:  number of MDTs to stripe a directory over\n"
656          "\tmdt_hash:   hash type of the striped directory. mdt types:\n"
657          "              all_char  (type 1)sum of characters % MDT_COUNT\n"
658          "              fnv_1a_64 (type 2)FNV-1a hash algorithm (default)\n"
659          "              crush     (type 3)CRUSH hash algorithm\n"
660          "\n"
661          "migrate file objects from one OST "
662          "layout\nto another (may be not safe with concurent writes).\n"
663          "usage: migrate [--stripe-count|-c STRIPE_COUNT]\n"
664          "               [--overstripe-count|-C STRIPE_COUNT]\n"
665          "               [--stripe-index|-i START_OST_INDEX]\n"
666          "               [--stripe-size|-S STRIPE_SIZE]\n"
667          "               [--pool|-p <pool_name>] [--ost|-o OST_INDICES]\n"
668          "               [--block|-b] [--non-block|-n] [--non-direct|-D]\n"
669          "               <file|directory>\n"
670          "\tstripe_count:   number of OSTs to stripe a file over\n"
671          "\t                Using -C instead of -c allows overstriping, which\n"
672          "\t                will place more than one stripe per OST if\n"
673          "\t                stripe_count is greater than the number of OSTs\n"
674          "\tstripe_ost_index: index of the first OST to stripe a file over\n"
675          "\tstripe_size:      number of bytes to store before moving to the next OST\n"
676          "\tpool_name:        name of the predefined pool of OSTs\n"
677          "\tost_indices:      OSTs to stripe over, in order\n"
678          "\tblock:        Block file access during data migration (default)\n"
679          "\tnon-block:    Abort migrations if concurrent access is detected\n"
680          "\tnon-direct:       do not use direct I/O to copy file contents.\n"},
681         {"mv", lfs_mv, 0,
682          "To move directories between MDTs. This command is deprecated, "
683          "use \"migrate\" instead.\n"
684          "usage: mv <directory|filename> [--mdt-index|-m MDT_INDEX] "
685          "[--verbose|-v]\n"},
686         {"ladvise", lfs_ladvise, 0,
687          "Provide servers with advice about access patterns for a file.\n"
688          "usage: ladvise [--advice|-a ADVICE] [--start|-s START[kMGT]]\n"
689          "               [--background|-b] [--unset|-u]\n\n"
690          "               {--end|-e END[kMGT]|--length|-l LENGTH[kMGT]}\n"
691          "               {[--mode|-m [READ,WRITE]}\n"
692          "               <file> ...\n"},
693         {"mirror", lfs_mirror, mirror_cmdlist,
694          "lfs commands used to manage files with mirrored components:\n"
695          "lfs mirror create - create a mirrored file or directory\n"
696          "lfs mirror extend - add mirror(s) to an existing file\n"
697          "lfs mirror split  - split a mirror from an existing mirrored file\n"
698          "lfs mirror resync - resynchronize out-of-sync mirrored file(s)\n"
699          "lfs mirror read   - read a mirror content of a mirrored file\n"
700          "lfs mirror write  - write to a mirror of a mirrored file\n"
701          "lfs mirror copy   - copy a mirror to other mirror(s) of a file\n"
702          "lfs mirror verify - verify mirrored file(s)\n"},
703         {"getsom", lfs_getsom, 0, "To list the SOM info for a given file.\n"
704          "usage: getsom [-s] [-b] [-f] <path>\n"
705          "\t-s: Only show the size value of the SOM data for a given file\n"
706          "\t-b: Only show the blocks value of the SOM data for a given file\n"
707          "\t-f: Only show the flags value of the SOM data for a given file\n"},
708         {"heat_get", lfs_heat_get, 0,
709          "To get heat of files.\n"
710          "usage: heat_get <file> ...\n"},
711         {"heat_set", lfs_heat_set, 0,
712          "To set heat flags of files.\n"
713          "usage: heat_set [--clear|-c] [--off|-o] [--on|-O] <file> ...\n"
714          "\t--clear|-c: Clear file heat for given files\n"
715          "\t--off|-o:   Turn off file heat for given files\n"
716          "\t--on|-O:    Turn on file heat for given files\n"},
717         {"pcc", lfs_pcc, pcc_cmdlist,
718          "lfs commands used to interact with PCC features:\n"
719          "lfs pcc attach - attach given files to Persistent Client Cache\n"
720          "lfs pcc attach_fid - attach given files into PCC by FID(s)\n"
721          "lfs pcc state  - display the PCC state for given files\n"
722          "lfs pcc detach - detach given files from Persistent Client Cache\n"
723          "lfs pcc detach_fid - detach given files from PCC by FID(s)\n"},
724         {"help", Parser_help, 0, "help"},
725         {"exit", Parser_quit, 0, "quit"},
726         {"quit", Parser_quit, 0, "quit"},
727         {"--version", Parser_version, 0,
728          "output build version of the utility and exit"},
729         {"--list-commands", lfs_list_commands, 0,
730          "list commands supported by the utility and exit"},
731         { 0, 0, 0, NULL }
732 };
733
734 static int check_hashtype(const char *hashtype)
735 {
736         int type_num = atoi(hashtype);
737         int i;
738
739         /* numeric hash type */
740         if (hashtype && strlen(hashtype) == 1 &&
741             (type_num > 0 && type_num < LMV_HASH_TYPE_MAX))
742                 return type_num;
743         /* string hash type */
744         for (i = LMV_HASH_TYPE_ALL_CHARS; i < LMV_HASH_TYPE_MAX; i++)
745                 if (strcmp(hashtype, mdt_hash_name[i]) == 0)
746                         return i;
747
748         return 0;
749 }
750
751 static uint32_t check_foreign_type_name(const char *foreign_type_name)
752 {
753         uint32_t i;
754
755         for (i = 0; i < LU_FOREIGN_TYPE_UNKNOWN; i++) {
756                 if (!lu_foreign_types[i].lft_name)
757                         break;
758                 if (strcmp(foreign_type_name,
759                            lu_foreign_types[i].lft_name) == 0)
760                         return lu_foreign_types[i].lft_type;
761         }
762
763         return LU_FOREIGN_TYPE_UNKNOWN;
764 }
765
766 static const char *error_loc = "syserror";
767
768 enum {
769         MIGRATION_NONBLOCK      = 0x0001,
770         MIGRATION_MIRROR        = 0x0002,
771         MIGRATION_NONDIRECT     = 0x0004,
772         MIGRATION_VERBOSE       = 0x0008,
773 };
774
775 static int
776 migrate_open_files(const char *name, __u64 migration_flags,
777                    const struct llapi_stripe_param *param,
778                    struct llapi_layout *layout, int *fd_src, int *fd_tgt)
779 {
780         int                      fd = -1;
781         int                      fdv = -1;
782         int                      rflags;
783         int                      mdt_index;
784         int                      random_value;
785         char                     parent[PATH_MAX];
786         char                     volatile_file[PATH_MAX];
787         char                    *ptr;
788         int                      rc;
789         struct stat              st;
790         struct stat              stv;
791
792         if (!param && !layout) {
793                 error_loc = "layout information";
794                 return -EINVAL;
795         }
796
797         /* search for file directory pathname */
798         if (strlen(name) > sizeof(parent) - 1) {
799                 error_loc = "source file name";
800                 return -ERANGE;
801         }
802
803         strncpy(parent, name, sizeof(parent));
804         ptr = strrchr(parent, '/');
805         if (!ptr) {
806                 if (!getcwd(parent, sizeof(parent))) {
807                         error_loc = "getcwd";
808                         return -errno;
809                 }
810         } else {
811                 if (ptr == parent) /* leading '/' */
812                         ptr = parent + 1;
813                 *ptr = '\0';
814         }
815
816         /* open file, direct io */
817         /* even if the file is only read, WR mode is nedeed to allow
818          * layout swap on fd
819          */
820         rflags = O_RDWR | O_NOATIME;
821         if (!(migration_flags & MIGRATION_NONDIRECT))
822                 rflags |= O_DIRECT;
823         fd = open(name, rflags);
824         if (fd < 0) {
825                 rc = -errno;
826                 error_loc = "cannot open source file";
827                 return rc;
828         }
829
830         rc = llapi_file_fget_mdtidx(fd, &mdt_index);
831         if (rc < 0) {
832                 error_loc = "cannot get MDT index";
833                 goto out;
834         }
835
836         do {
837                 int open_flags = O_WRONLY | O_CREAT | O_EXCL | O_NOFOLLOW;
838                 mode_t open_mode = S_IRUSR | S_IWUSR;
839
840                 random_value = random();
841                 rc = snprintf(volatile_file, sizeof(volatile_file),
842                               "%s/%s:%.4X:%.4X:fd=%.2d", parent,
843                               LUSTRE_VOLATILE_HDR, mdt_index,
844                               random_value, fd);
845                 if (rc >= sizeof(volatile_file)) {
846                         rc = -ENAMETOOLONG;
847                         break;
848                 }
849
850                 /* create, open a volatile file, use caching (ie no directio) */
851                 if (layout) {
852                         /* Returns -1 and sets errno on error: */
853                         fdv = llapi_layout_file_open(volatile_file, open_flags,
854                                                      open_mode, layout);
855                         if (fdv < 0)
856                                 fdv = -errno;
857                 } else {
858                         /* Does the right thing on error: */
859                         fdv = llapi_file_open_param(volatile_file, open_flags,
860                                                     open_mode, param);
861                 }
862         } while (fdv < 0 && (rc = fdv) == -EEXIST);
863
864         if (rc < 0) {
865                 error_loc = "cannot create volatile file";
866                 goto out;
867         }
868
869         /*
870          * In case the MDT does not support creation of volatile files
871          * we should try to unlink it.
872          */
873         (void)unlink(volatile_file);
874
875         /*
876          * Not-owner (root?) special case.
877          * Need to set owner/group of volatile file like original.
878          * This will allow to pass related check during layout_swap.
879          */
880         rc = fstat(fd, &st);
881         if (rc != 0) {
882                 rc = -errno;
883                 error_loc = "cannot stat source file";
884                 goto out;
885         }
886
887         rc = fstat(fdv, &stv);
888         if (rc != 0) {
889                 rc = -errno;
890                 error_loc = "cannot stat volatile";
891                 goto out;
892         }
893
894         if (st.st_uid != stv.st_uid || st.st_gid != stv.st_gid) {
895                 rc = fchown(fdv, st.st_uid, st.st_gid);
896                 if (rc != 0) {
897                         rc = -errno;
898                         error_loc = "cannot change ownwership of volatile";
899                         goto out;
900                 }
901         }
902
903 out:
904         if (rc < 0) {
905                 if (fd > 0)
906                         close(fd);
907                 if (fdv > 0)
908                         close(fdv);
909         } else {
910                 *fd_src = fd;
911                 *fd_tgt = fdv;
912                 error_loc = NULL;
913         }
914         return rc;
915 }
916
917 static int migrate_copy_data(int fd_src, int fd_dst, int (*check_file)(int))
918 {
919         struct llapi_layout *layout;
920         size_t buf_size = 4 * 1024 * 1024;
921         void *buf = NULL;
922         off_t pos = 0;
923         off_t data_end = 0;
924         size_t page_size = sysconf(_SC_PAGESIZE);
925         bool sparse;
926         int rc;
927
928         layout = llapi_layout_get_by_fd(fd_src, 0);
929         if (layout) {
930                 uint64_t stripe_size;
931
932                 rc = llapi_layout_stripe_size_get(layout, &stripe_size);
933                 if (rc == 0)
934                         buf_size = stripe_size;
935
936                 llapi_layout_free(layout);
937         }
938
939         /* Use a page-aligned buffer for direct I/O */
940         rc = posix_memalign(&buf, page_size, buf_size);
941         if (rc != 0)
942                 return -rc;
943
944         sparse = llapi_file_is_sparse(fd_src);
945         if (sparse) {
946                 rc = ftruncate(fd_dst, pos);
947                 if (rc < 0) {
948                         rc = -errno;
949                         return rc;
950                 }
951         }
952
953         while (1) {
954                 off_t data_off;
955                 size_t to_read, to_write;
956                 ssize_t rsize;
957
958                 if (sparse && pos >= data_end) {
959                         size_t data_size;
960
961                         data_off = llapi_data_seek(fd_src, pos, &data_size);
962                         if (data_off < 0) {
963                                 /* Non-fatal, switch to full copy */
964                                 sparse = false;
965                                 continue;
966                         }
967                         /* hole at the end of file, truncate up to it */
968                         if (!data_size) {
969                                 rc = ftruncate(fd_dst, data_off);
970                                 if (rc < 0)
971                                         goto out;
972                         }
973                         pos = data_off & ~(page_size - 1);
974                         data_end = data_off + data_size;
975                         to_read = ((data_end - pos - 1) | (page_size - 1)) + 1;
976                         to_read = MIN(to_read, buf_size);
977                 } else {
978                         to_read = buf_size;
979                 }
980
981                 if (check_file) {
982                         rc = check_file(fd_src);
983                         if (rc < 0)
984                                 goto out;
985                 }
986
987                 rsize = pread(fd_src, buf, to_read, pos);
988                 if (rsize < 0) {
989                         rc = -errno;
990                         goto out;
991                 }
992                 /* EOF */
993                 if (rsize == 0)
994                         break;
995
996                 to_write = rsize;
997                 while (to_write > 0) {
998                         ssize_t written;
999
1000                         written = pwrite(fd_dst, buf, to_write, pos);
1001                         if (written < 0) {
1002                                 rc = -errno;
1003                                 goto out;
1004                         }
1005                         pos += written;
1006                         to_write -= written;
1007                 }
1008                 if (rc || rsize < to_read)
1009                         break;
1010         }
1011
1012         rc = fsync(fd_dst);
1013         if (rc < 0)
1014                 rc = -errno;
1015 out:
1016         /* Try to avoid page cache pollution after migration. */
1017         (void)posix_fadvise(fd_src, 0, 0, POSIX_FADV_DONTNEED);
1018         (void)posix_fadvise(fd_dst, 0, 0, POSIX_FADV_DONTNEED);
1019
1020         free(buf);
1021         return rc;
1022 }
1023
1024 static int migrate_set_timestamps(int fd, const struct stat *st)
1025 {
1026         struct timeval tv[2] = {
1027                 {.tv_sec = st->st_atime},
1028                 {.tv_sec = st->st_mtime}
1029         };
1030
1031         return futimes(fd, tv);
1032 }
1033
1034 static int migrate_block(int fd, int fdv)
1035 {
1036         struct stat st;
1037         __u64   dv1;
1038         int     gid;
1039         int     rc;
1040         int     rc2;
1041
1042         rc = fstat(fd, &st);
1043         if (rc < 0) {
1044                 error_loc = "cannot stat source file";
1045                 return -errno;
1046         }
1047
1048         rc = llapi_get_data_version(fd, &dv1, LL_DV_RD_FLUSH);
1049         if (rc < 0) {
1050                 error_loc = "cannot get dataversion";
1051                 return rc;
1052         }
1053
1054         do
1055                 gid = random();
1056         while (gid == 0);
1057
1058         /*
1059          * The grouplock blocks all concurrent accesses to the file.
1060          * It has to be taken after llapi_get_data_version as it would
1061          * block it too.
1062          */
1063         rc = llapi_group_lock(fd, gid);
1064         if (rc < 0) {
1065                 error_loc = "cannot get group lock";
1066                 return rc;
1067         }
1068
1069         rc = migrate_copy_data(fd, fdv, NULL);
1070         if (rc < 0) {
1071                 error_loc = "data copy failed";
1072                 goto out_unlock;
1073         }
1074
1075         /* Make sure we keep original atime/mtime values */
1076         rc = migrate_set_timestamps(fdv, &st);
1077         if (rc < 0) {
1078                 error_loc = "set target file timestamp failed";
1079                 goto out_unlock;
1080         }
1081
1082         /*
1083          * swap layouts
1084          * for a migration we need to check data version on file did
1085          * not change.
1086          *
1087          * Pass in gid=0 since we already own grouplock.
1088          */
1089         rc = llapi_fswap_layouts_grouplock(fd, fdv, dv1, 0, 0,
1090                                            SWAP_LAYOUTS_CHECK_DV1);
1091         if (rc == -EAGAIN) {
1092                 error_loc = "file changed";
1093                 goto out_unlock;
1094         } else if (rc < 0) {
1095                 error_loc = "cannot swap layout";
1096                 goto out_unlock;
1097         }
1098
1099 out_unlock:
1100         rc2 = llapi_group_unlock(fd, gid);
1101         if (rc2 < 0 && rc == 0) {
1102                 error_loc = "unlock group lock";
1103                 rc = rc2;
1104         }
1105
1106         return rc;
1107 }
1108
1109 /**
1110  * Internal helper for migrate_copy_data(). Check lease and report error if
1111  * need be.
1112  *
1113  * \param[in]  fd           File descriptor on which to check the lease.
1114  *
1115  * \retval 0       Migration can keep on going.
1116  * \retval -errno  Error occurred, abort migration.
1117  */
1118 static int check_lease(int fd)
1119 {
1120         int rc;
1121
1122         rc = llapi_lease_check(fd);
1123         if (rc > 0)
1124                 return 0; /* llapi_check_lease returns > 0 on success. */
1125
1126         return -EBUSY;
1127 }
1128
1129 static int migrate_nonblock(int fd, int fdv)
1130 {
1131         struct stat st;
1132         __u64   dv1;
1133         __u64   dv2;
1134         int     rc;
1135
1136         rc = fstat(fd, &st);
1137         if (rc < 0) {
1138                 error_loc = "cannot stat source file";
1139                 return -errno;
1140         }
1141
1142         rc = llapi_get_data_version(fd, &dv1, LL_DV_RD_FLUSH);
1143         if (rc < 0) {
1144                 error_loc = "cannot get data version";
1145                 return rc;
1146         }
1147
1148         rc = migrate_copy_data(fd, fdv, check_lease);
1149         if (rc < 0) {
1150                 error_loc = "data copy failed";
1151                 return rc;
1152         }
1153
1154         rc = llapi_get_data_version(fd, &dv2, LL_DV_RD_FLUSH);
1155         if (rc != 0) {
1156                 error_loc = "cannot get data version";
1157                 return rc;
1158         }
1159
1160         if (dv1 != dv2) {
1161                 rc = -EAGAIN;
1162                 error_loc = "source file changed";
1163                 return rc;
1164         }
1165
1166         /* Make sure we keep original atime/mtime values */
1167         rc = migrate_set_timestamps(fdv, &st);
1168         if (rc < 0) {
1169                 error_loc = "set target file timestamp failed";
1170                 return -errno;
1171         }
1172         return 0;
1173 }
1174
1175 static
1176 int lfs_layout_compid_by_pool(char *fname, const char *pool, int *comp_id)
1177 {
1178         struct pool_to_id_cbdata data = { .pool = pool };
1179         struct llapi_layout *layout = NULL;
1180         int rc;
1181
1182         layout = llapi_layout_get_by_path(fname, 0);
1183         if (!layout) {
1184                 fprintf(stderr,
1185                         "error %s: file '%s' couldn't get layout: rc=%d\n",
1186                         progname, fname, errno);
1187                 rc = -errno;
1188                 goto free_layout;
1189         }
1190         rc = llapi_layout_sanity(layout, fname, false, true);
1191         if (rc < 0) {
1192                 llapi_layout_sanity_perror(errno);
1193                 goto free_layout;
1194         }
1195         rc = llapi_layout_comp_iterate(layout, find_comp_id_by_pool, &data);
1196         if (rc < 0)
1197                 goto free_layout;
1198
1199         *comp_id = data.id;
1200         rc = 0;
1201
1202 free_layout:
1203         if (layout)
1204                 llapi_layout_free(layout);
1205         return rc;
1206 }
1207
1208 static int lfs_component_set(char *fname, int comp_id, const char *pool,
1209                              __u32 flags, __u32 neg_flags)
1210 {
1211         __u32 ids[2];
1212         __u32 flags_array[2];
1213         size_t count = 0;
1214         int rc;
1215
1216         if (!comp_id) {
1217                 if (pool == NULL) {
1218                         fprintf(stderr,
1219                                 "error %s: neither component id nor pool is specified\n",
1220                                 progname);
1221                         return -EINVAL;
1222                 }
1223                 rc = lfs_layout_compid_by_pool(fname, pool, &comp_id);
1224                 if (rc)
1225                         return rc;
1226         }
1227
1228         if (flags) {
1229                 ids[count] = comp_id;
1230                 flags_array[count] = flags;
1231                 ++count;
1232         }
1233
1234         if (neg_flags) {
1235                 if (neg_flags & LCME_FL_STALE) {
1236                         fprintf(stderr,
1237                                 "%s: cannot clear 'stale' flags from component. Please use lfs-mirror-resync(1) instead\n",
1238                                 progname);
1239                         return -EINVAL;
1240                 }
1241
1242                 ids[count] = comp_id;
1243                 flags_array[count] = neg_flags | LCME_FL_NEG;
1244                 ++count;
1245         }
1246
1247         rc = llapi_layout_file_comp_set(fname, ids, flags_array, count);
1248         if (rc) {
1249                 if (errno == EUCLEAN) {
1250                         rc = -errno;
1251                         fprintf(stderr,
1252                                 "%s: cannot set 'stale' flag on component '%#x' of the last non-stale mirror of '%s'\n",
1253                                 progname, comp_id, fname);
1254                 } else {
1255                         fprintf(stderr,
1256                                 "%s: cannot change the flags of component '%#x' of file '%s': %x / ^(%x)\n",
1257                                 progname, comp_id, fname, flags, neg_flags);
1258                 }
1259         }
1260
1261         return rc;
1262 }
1263
1264 static int lfs_component_del(char *fname, __u32 comp_id,
1265                              __u32 flags, __u32 neg_flags)
1266 {
1267         int     rc = 0;
1268
1269         if (flags && neg_flags)
1270                 return -EINVAL;
1271
1272         if (!flags && neg_flags)
1273                 flags = neg_flags | LCME_FL_NEG;
1274
1275         if ((flags && comp_id) || (!flags && !comp_id))
1276                 return -EINVAL;
1277
1278         if (flags) {
1279                 if (flags & ~LCME_KNOWN_FLAGS) {
1280                         fprintf(stderr,
1281                                 "%s setstripe: unknown flags %#x\n",
1282                                 progname, flags);
1283                         return -EINVAL;
1284                 }
1285         } else if (comp_id > LCME_ID_MAX) {
1286                 fprintf(stderr, "%s setstripe: invalid component id %u\n",
1287                         progname, comp_id);
1288                 return -EINVAL;
1289         }
1290
1291         rc = llapi_layout_file_comp_del(fname, comp_id, flags);
1292         if (rc)
1293                 fprintf(stderr,
1294                         "%s setstripe: cannot delete component %#x from '%s': %s\n",
1295                         progname, comp_id, fname, strerror(errno));
1296         return rc;
1297 }
1298
1299 static int lfs_component_add(char *fname, struct llapi_layout *layout)
1300 {
1301         int     rc;
1302
1303         if (!layout)
1304                 return -EINVAL;
1305
1306         rc = llapi_layout_file_comp_add(fname, layout);
1307         if (rc)
1308                 fprintf(stderr, "Add layout component(s) to %s failed. %s\n",
1309                         fname, strerror(errno));
1310         return rc;
1311 }
1312
1313 static int lfs_component_create(char *fname, int open_flags, mode_t open_mode,
1314                                 struct llapi_layout *layout)
1315 {
1316         struct stat     st;
1317         int     fd;
1318
1319         if (!layout)
1320                 return -EINVAL;
1321
1322         fd = lstat(fname, &st);
1323         if (fd == 0 && S_ISDIR(st.st_mode))
1324                 open_flags = O_DIRECTORY | O_RDONLY;
1325
1326         fd = llapi_layout_file_open(fname, open_flags, open_mode, layout);
1327         if (fd < 0)
1328                 fprintf(stderr, "%s: cannot %s '%s': %s\n", progname,
1329                         S_ISDIR(st.st_mode) ?
1330                                 "set default composite layout for" :
1331                                 "create composite file",
1332                         fname, strerror(errno));
1333         return fd;
1334 }
1335
1336 static int lfs_migrate(char *name, __u64 migration_flags,
1337                        struct llapi_stripe_param *param,
1338                        struct llapi_layout *layout)
1339 {
1340         struct llapi_layout *existing;
1341         uint64_t dom_new, dom_cur;
1342         int fd = -1;
1343         int fdv = -1;
1344         int rc;
1345
1346         rc = migrate_open_files(name, migration_flags, param, layout,
1347                                 &fd, &fdv);
1348         if (rc < 0)
1349                 goto out;
1350
1351         rc = llapi_layout_dom_size(layout, &dom_new);
1352         if (rc) {
1353                 error_loc = "cannot get new layout DoM size";
1354                 goto out;
1355         }
1356         /* special case for migration to DOM layout*/
1357         existing = llapi_layout_get_by_fd(fd, 0);
1358         if (!existing) {
1359                 error_loc = "cannot get existing layout";
1360                 goto out;
1361         }
1362
1363         rc = llapi_layout_dom_size(existing, &dom_cur);
1364         if (rc) {
1365                 error_loc = "cannot get current layout DoM size";
1366                 goto out;
1367         }
1368
1369         /*
1370          * if file has DoM layout already then migration is possible to
1371          * the new layout with the same DoM component via swap layout,
1372          * if new layout used bigger DOM size, then mirroring is used
1373          */
1374         if (dom_new > dom_cur) {
1375                 rc = lfs_migrate_to_dom(fd, fdv, name, migration_flags, param,
1376                                         layout);
1377                 if (rc)
1378                         error_loc = "cannot migrate to DOM layout";
1379                 goto out_closed;
1380         }
1381
1382         if (!(migration_flags & MIGRATION_NONBLOCK)) {
1383                 /*
1384                  * Blocking mode (forced if servers do not support file lease).
1385                  * It is also the default mode, since we cannot distinguish
1386                  * between a broken lease and a server that does not support
1387                  * atomic swap/close (LU-6785)
1388                  */
1389                 rc = migrate_block(fd, fdv);
1390                 goto out;
1391         }
1392
1393         rc = llapi_lease_acquire(fd, LL_LEASE_RDLCK);
1394         if (rc < 0) {
1395                 error_loc = "cannot get lease";
1396                 goto out;
1397         }
1398
1399         rc = migrate_nonblock(fd, fdv);
1400         if (rc < 0) {
1401                 llapi_lease_release(fd);
1402                 goto out;
1403         }
1404
1405         /*
1406          * Atomically put lease, swap layouts and close.
1407          * for a migration we need to check data version on file did
1408          * not change.
1409          */
1410         rc = llapi_fswap_layouts(fd, fdv, 0, 0, SWAP_LAYOUTS_CLOSE);
1411         if (rc < 0) {
1412                 error_loc = "cannot swap layout";
1413                 goto out;
1414         }
1415
1416 out:
1417         if (fd >= 0)
1418                 close(fd);
1419
1420         if (fdv >= 0)
1421                 close(fdv);
1422 out_closed:
1423         if (rc < 0)
1424                 fprintf(stderr, "error: %s: %s: %s: %s\n",
1425                         progname, name, error_loc, strerror(-rc));
1426         else if (migration_flags & MIGRATION_VERBOSE)
1427                 printf("%s\n", name);
1428
1429         return rc;
1430 }
1431
1432 static int comp_str2flags(char *string, __u32 *flags, __u32 *neg_flags)
1433 {
1434         char *name;
1435         char *dup_string = NULL;
1436         int rc = 0;
1437
1438         *flags = 0;
1439         *neg_flags = 0;
1440
1441         if (!string || !string[0])
1442                 return -EINVAL;
1443
1444         dup_string = strdup(string);
1445         if (!dup_string) {
1446                 llapi_printf(LLAPI_MSG_ERROR,
1447                              "%s: insufficient memory\n",
1448                              progname);
1449                 return -ENOMEM;
1450         }
1451
1452         for (name = strtok(dup_string, ","); name; name = strtok(NULL, ",")) {
1453                 bool found = false;
1454                 int i;
1455
1456                 for (i = 0; i < ARRAY_SIZE(comp_flags_table); i++) {
1457                         __u32 comp_flag = comp_flags_table[i].cfn_flag;
1458                         const char *comp_name = comp_flags_table[i].cfn_name;
1459
1460                         if (strcmp(name, comp_name) == 0) {
1461                                 *flags |= comp_flag;
1462                                 found = true;
1463                         } else if (strncmp(name, "^", 1) == 0 &&
1464                                    strcmp(name + 1, comp_name) == 0) {
1465                                 *neg_flags |= comp_flag;
1466                                 found = true;
1467                         }
1468                 }
1469                 if (!found) {
1470                         llapi_printf(LLAPI_MSG_ERROR,
1471                                      "%s: component flag '%s' not supported\n",
1472                                      progname, name);
1473                         rc = -EINVAL;
1474                         goto out_free;
1475                 }
1476         }
1477
1478         if (!*flags && !*neg_flags)
1479                 rc = -EINVAL;
1480
1481         /* don't allow to set and exclude the same flag */
1482         if (*flags & *neg_flags)
1483                 rc = -EINVAL;
1484
1485 out_free:
1486         free(dup_string);
1487         return rc;
1488 }
1489
1490 static int mdthash_input(char *string, __u32 *inflags,
1491                          __u32 *exflags, __u32 *type)
1492 {
1493         char *name;
1494         struct mhf_list {
1495                 char *name;
1496                 __u32 flag;
1497         } mhflist[] = {
1498                 {"migrating", LMV_HASH_FLAG_MIGRATION},
1499                 {"badtype", LMV_HASH_FLAG_BAD_TYPE},
1500                 {"lostlmv", LMV_HASH_FLAG_LOST_LMV},
1501         };
1502
1503         if (string == NULL)
1504                 return -EINVAL;
1505
1506         *inflags = 0;
1507         *exflags = 0;
1508         *type = 0;
1509         for (name = strtok(string, ","); name; name = strtok(NULL, ",")) {
1510                 bool found = false;
1511                 int i;
1512
1513                 for (i = 0; i < ARRAY_SIZE(mhflist); i++) {
1514                         if (strcmp(name, mhflist[i].name) == 0 ||
1515                             name[0] == mhflist[i].name[0]) {
1516                                 *inflags |= mhflist[i].flag;
1517                                 found = true;
1518                         } else if (name[0] == '^' &&
1519                                    (strcmp(name + 1, mhflist[i].name) == 0 ||
1520                                     name[1] == mhflist[i].name[0])) {
1521                                 *exflags |= mhflist[i].flag;
1522                                 found = true;
1523                         }
1524                 }
1525                 if (!found) {
1526                         i = check_hashtype(name);
1527                         if (i > 0) {
1528                                 *type |= 1 << i;
1529                                 continue;
1530                         }
1531                         llapi_printf(LLAPI_MSG_ERROR,
1532                                      "%s: invalid mdt_hash value '%s'\n",
1533                                      progname, name);
1534                         return -EINVAL;
1535                 }
1536         }
1537
1538         /* don't allow to include and exclude the same flag */
1539         if (*inflags & *exflags) {
1540                 llapi_printf(LLAPI_MSG_ERROR,
1541                              "%s: include and exclude same flag '%s'\n",
1542                              progname, string);
1543                 return -EINVAL;
1544         }
1545
1546         return 0;
1547 }
1548
1549 static int mirror_str2state(char *string, __u16 *state, __u16 *neg_state)
1550 {
1551         if (!string)
1552                 return -EINVAL;
1553
1554         *state = 0;
1555         *neg_state = 0;
1556
1557         if (strncmp(string, "^", 1) == 0) {
1558                 *neg_state = llapi_layout_string_flags(string + 1);
1559                 if (*neg_state != 0)
1560                         return 0;
1561         } else {
1562                 *state = llapi_layout_string_flags(string);
1563                 if (*state != 0)
1564                         return 0;
1565         }
1566
1567         llapi_printf(LLAPI_MSG_ERROR,
1568                      "%s: mirrored file state '%s' not supported\n",
1569                      progname, string);
1570         return -EINVAL;
1571 }
1572
1573 /**
1574  * struct mirror_args - Command-line arguments for mirror(s).
1575  * @m_count:  Number of mirrors to be created with this layout.
1576  * @m_flags:  Mirror level flags, only 'prefer' is supported.
1577  * @m_layout: Mirror layout.
1578  * @m_file:   A victim file. Its layout will be split and used as a mirror.
1579  * @m_next:   Point to the next node of the list.
1580  *
1581  * Command-line arguments for mirror(s) will be parsed and stored in
1582  * a linked list that consists of this structure.
1583  */
1584 struct mirror_args {
1585         __u32                   m_count;
1586         __u32                   m_flags;
1587         struct llapi_layout     *m_layout;
1588         const char              *m_file;
1589         struct mirror_args      *m_next;
1590         bool                    m_inherit;
1591 };
1592
1593 /**
1594  * enum mirror_flags - Flags for extending a mirrored file.
1595  * @MF_NO_VERIFY: Indicates not to verify the mirror(s) from victim file(s)
1596  *             in case the victim file(s) contains the same data as the
1597  *             original mirrored file.
1598  * @MF_DESTROY: Indicates to delete the mirror from the mirrored file.
1599  * @MF_COMP_ID: specified component id instead of mirror id
1600  *
1601  * Flags for extending a mirrored file.
1602  */
1603 enum mirror_flags {
1604         MF_NO_VERIFY    = 0x1,
1605         MF_DESTROY      = 0x2,
1606         MF_COMP_ID      = 0x4,
1607         MF_COMP_POOL    = 0x8,
1608 };
1609
1610 /**
1611  * mirror_create_sanity_check() - Check mirror list.
1612  * @list:  A linked list that stores the mirror arguments.
1613  *
1614  * This function does a sanity check on @list for creating
1615  * a mirrored file.
1616  *
1617  * Return: 0 on success or a negative error code on failure.
1618  */
1619 static int mirror_create_sanity_check(const char *fname,
1620                                       struct mirror_args *list,
1621                                       bool check_fname)
1622 {
1623         int rc = 0;
1624         bool has_m_file = false;
1625         bool has_m_layout = false;
1626
1627         if (!list)
1628                 return -EINVAL;
1629
1630         if (fname && check_fname) {
1631                 struct llapi_layout *layout;
1632
1633                 layout = llapi_layout_get_by_path(fname, 0);
1634                 if (!layout) {
1635                         fprintf(stderr,
1636                                 "error: %s: file '%s' couldn't get layout\n",
1637                                 progname, fname);
1638                         return -ENODATA;
1639                 }
1640
1641                 rc = llapi_layout_sanity(layout, fname, false, true);
1642
1643                 llapi_layout_free(layout);
1644
1645                 if (rc) {
1646                         llapi_layout_sanity_perror(rc);
1647                         return rc;
1648                 }
1649         }
1650
1651         while (list) {
1652                 if (list->m_file) {
1653                         has_m_file = true;
1654                         llapi_layout_free(list->m_layout);
1655
1656                         list->m_layout =
1657                                 llapi_layout_get_by_path(list->m_file, 0);
1658                         if (!list->m_layout) {
1659                                 fprintf(stderr,
1660                                         "error: %s: file '%s' has no layout\n",
1661                                         progname, list->m_file);
1662                                 return -ENODATA;
1663                         }
1664                 } else {
1665                         has_m_layout = true;
1666                         if (!list->m_layout) {
1667                                 fprintf(stderr, "error: %s: no mirror layout\n",
1668                                         progname);
1669                                 return -EINVAL;
1670                         }
1671                 }
1672
1673                 rc = llapi_layout_sanity(list->m_layout, fname, false, true);
1674                 if (rc) {
1675                         llapi_layout_sanity_perror(rc);
1676                         return rc;
1677                 }
1678
1679                 list = list->m_next;
1680         }
1681
1682         if (has_m_file && has_m_layout) {
1683                 fprintf(stderr,
1684                         "error: %s: -f <victim_file> option should not be specified with setstripe options\n",
1685                         progname);
1686                 return -EINVAL;
1687         }
1688
1689         return 0;
1690 }
1691
1692 static int mirror_set_flags(struct llapi_layout *layout, void *cbdata)
1693 {
1694         __u32 mirror_flags = *(__u32 *)cbdata;
1695         uint32_t flags;
1696         int rc;
1697
1698         rc = llapi_layout_comp_flags_get(layout, &flags);
1699         if (rc < 0)
1700                 return rc;
1701
1702         if (!flags) {
1703                 rc = llapi_layout_comp_flags_set(layout, mirror_flags);
1704                 if (rc)
1705                         return rc;
1706         }
1707
1708         return LLAPI_LAYOUT_ITER_CONT;
1709 }
1710
1711 /**
1712  * mirror_create() - Create a mirrored file.
1713  * @fname:        The file to be created.
1714  * @mirror_list:  A linked list that stores the mirror arguments.
1715  *
1716  * This function creates a mirrored file @fname with the mirror(s)
1717  * from @mirror_list.
1718  *
1719  * Return: 0 on success or a negative error code on failure.
1720  */
1721 static int mirror_create(char *fname, struct mirror_args *mirror_list)
1722 {
1723         struct llapi_layout *layout = NULL;
1724         struct mirror_args *cur_mirror = NULL;
1725         uint16_t mirror_count = 0;
1726         int i = 0;
1727         int rc = 0;
1728
1729         rc = mirror_create_sanity_check(fname, mirror_list, false);
1730         if (rc)
1731                 return rc;
1732
1733         cur_mirror = mirror_list;
1734         while (cur_mirror) {
1735                 rc = llapi_layout_comp_iterate(cur_mirror->m_layout,
1736                                                mirror_set_flags,
1737                                                &cur_mirror->m_flags);
1738                 if (rc) {
1739                         rc = -errno;
1740                         fprintf(stderr, "%s: failed to set mirror flags\n",
1741                                 progname);
1742                         goto error;
1743                 }
1744
1745                 for (i = 0; i < cur_mirror->m_count; i++) {
1746                         rc = llapi_layout_merge(&layout, cur_mirror->m_layout);
1747                         if (rc) {
1748                                 rc = -errno;
1749                                 fprintf(stderr,
1750                                         "error: %s: merge layout failed: %s\n",
1751                                         progname, strerror(errno));
1752                                 goto error;
1753                         }
1754                 }
1755                 mirror_count += cur_mirror->m_count;
1756                 cur_mirror = cur_mirror->m_next;
1757         }
1758
1759         if (!layout) {
1760                 fprintf(stderr, "error: %s: layout is NULL\n", progname);
1761                 return -EINVAL;
1762         }
1763
1764         rc = llapi_layout_mirror_count_set(layout, mirror_count);
1765         if (rc) {
1766                 rc = -errno;
1767                 fprintf(stderr, "error: %s: set mirror count failed: %s\n",
1768                         progname, strerror(errno));
1769                 goto error;
1770         }
1771
1772         rc = lfs_component_create(fname, O_CREAT | O_WRONLY, 0666,
1773                                   layout);
1774         if (rc >= 0) {
1775                 close(rc);
1776                 rc = 0;
1777         }
1778
1779 error:
1780         llapi_layout_free(layout);
1781         return rc;
1782 }
1783
1784 /**
1785  * Compare files and check lease on @fd.
1786  *
1787  * \retval bytes number of bytes are the same
1788  */
1789 static ssize_t mirror_file_compare(int fd, int fdv)
1790 {
1791         const size_t buflen = 4 * 1024 * 1024; /* 4M */
1792         void *buf;
1793         ssize_t bytes_done = 0;
1794         ssize_t bytes_read = 0;
1795
1796         buf = malloc(buflen * 2);
1797         if (!buf)
1798                 return -ENOMEM;
1799
1800         while (1) {
1801                 if (!llapi_lease_check(fd)) {
1802                         bytes_done = -EBUSY;
1803                         break;
1804                 }
1805
1806                 bytes_read = read(fd, buf, buflen);
1807                 if (bytes_read <= 0)
1808                         break;
1809
1810                 if (bytes_read != read(fdv, buf + buflen, buflen))
1811                         break;
1812
1813                 /*
1814                  * XXX: should compute the checksum on each buffer and then
1815                  * compare checksum to avoid cache collision
1816                  */
1817                 if (memcmp(buf, buf + buflen, bytes_read))
1818                         break;
1819
1820                 bytes_done += bytes_read;
1821         }
1822
1823         free(buf);
1824
1825         return bytes_done;
1826 }
1827
1828 static int mirror_extend_file(const char *fname, const char *victim_file,
1829                               enum mirror_flags mirror_flags)
1830 {
1831         int fd = -1;
1832         int fdv = -1;
1833         struct stat stbuf;
1834         struct stat stbuf_v;
1835         struct ll_ioc_lease *data = NULL;
1836         int rc;
1837
1838         fd = open(fname, O_RDWR);
1839         if (fd < 0) {
1840                 error_loc = "open source file";
1841                 rc = -errno;
1842                 goto out;
1843         }
1844
1845         fdv = open(victim_file, O_RDWR);
1846         if (fdv < 0) {
1847                 error_loc = "open target file";
1848                 rc = -errno;
1849                 goto out;
1850         }
1851
1852         if (fstat(fd, &stbuf) || fstat(fdv, &stbuf_v)) {
1853                 error_loc = "stat source or target file";
1854                 rc = -errno;
1855                 goto out;
1856         }
1857
1858         if (stbuf.st_dev != stbuf_v.st_dev) {
1859                 error_loc = "stat source and target file";
1860                 rc = -EXDEV;
1861                 goto out;
1862         }
1863
1864         /* mirrors should be of the same size */
1865         if (stbuf.st_size != stbuf_v.st_size) {
1866                 error_loc = "file sizes don't match";
1867                 rc = -EINVAL;
1868                 goto out;
1869         }
1870
1871         rc = llapi_lease_acquire(fd, LL_LEASE_RDLCK);
1872         if (rc < 0) {
1873                 error_loc = "cannot get lease";
1874                 goto out;
1875         }
1876
1877         if (!(mirror_flags & MF_NO_VERIFY)) {
1878                 ssize_t ret;
1879                 /* mirrors should have the same contents */
1880                 ret = mirror_file_compare(fd, fdv);
1881                 if (ret != stbuf.st_size) {
1882                         error_loc = "file busy or contents don't match";
1883                         rc = ret < 0 ? ret : -EINVAL;
1884                         goto out;
1885                 }
1886         }
1887
1888         /* Get rid of caching pages from clients */
1889         rc = llapi_file_flush(fd);
1890         if (rc < 0) {
1891                 error_loc = "cannot get data version";
1892                 goto out;
1893         }
1894
1895         rc = llapi_file_flush(fdv);
1896         if (rc < 0) {
1897                 error_loc = "cannot get data version";
1898                 goto out;
1899         }
1900
1901         rc = migrate_set_timestamps(fd, &stbuf);
1902         if (rc < 0) {
1903                 error_loc = "cannot set source file timestamp";
1904                 goto out;
1905         }
1906
1907         /* Atomically put lease, merge layouts and close. */
1908         data = calloc(1, offsetof(typeof(*data), lil_ids[1]));
1909         if (!data) {
1910                 error_loc = "memory allocation";
1911                 goto out;
1912         }
1913         data->lil_mode = LL_LEASE_UNLCK;
1914         data->lil_flags = LL_LEASE_LAYOUT_MERGE;
1915         data->lil_count = 1;
1916         data->lil_ids[0] = fdv;
1917         rc = llapi_lease_set(fd, data);
1918         if (rc < 0) {
1919                 error_loc = "cannot merge layout";
1920                 goto out;
1921         } else if (rc == 0) {
1922                 rc = -EBUSY;
1923                 error_loc = "lost lease lock";
1924                 goto out;
1925         }
1926         rc = 0;
1927
1928 out:
1929         if (data)
1930                 free(data);
1931         if (fd >= 0)
1932                 close(fd);
1933         if (fdv >= 0)
1934                 close(fdv);
1935         if (!rc)
1936                 (void) unlink(victim_file);
1937         if (rc < 0)
1938                 fprintf(stderr, "error: %s: %s: %s: %s\n",
1939                         progname, fname, error_loc, strerror(-rc));
1940         return rc;
1941 }
1942
1943 static int mirror_extend_layout(char *name, struct llapi_layout *m_layout,
1944                                 bool inherit, uint32_t flags)
1945 {
1946         struct llapi_layout *f_layout = NULL;
1947         struct ll_ioc_lease *data = NULL;
1948         struct stat st;
1949         int fd = -1;
1950         int fdv = -1;
1951         int rc = 0;
1952
1953         if (inherit) {
1954                 f_layout = llapi_layout_get_by_path(name, 0);
1955                 if (!f_layout) {
1956                         rc = -EINVAL;
1957                         fprintf(stderr, "%s: cannot get layout\n", progname);
1958                         goto out;
1959                 }
1960                 rc = llapi_layout_get_last_init_comp(f_layout);
1961                 if (rc) {
1962                         fprintf(stderr, "%s: cannot get the last init comp\n",
1963                                 progname);
1964                         goto out;
1965                 }
1966                 rc = llapi_layout_mirror_inherit(f_layout, m_layout);
1967                 if (rc) {
1968                         fprintf(stderr,
1969                                 "%s: cannot inherit from the last init comp\n",
1970                                 progname);
1971                         goto out;
1972                 }
1973         }
1974         llapi_layout_comp_flags_set(m_layout, flags);
1975         rc = migrate_open_files(name, MIGRATION_NONDIRECT, NULL, m_layout, &fd,
1976                                 &fdv);
1977         if (rc < 0)
1978                 goto out;
1979
1980         rc = llapi_lease_acquire(fd, LL_LEASE_RDLCK);
1981         if (rc < 0) {
1982                 error_loc = "cannot get lease";
1983                 goto out;
1984         }
1985
1986         rc = fstat(fd, &st);
1987         if (rc < 0) {
1988                 error_loc = "cannot stat source file";
1989                 goto out;
1990         }
1991
1992         rc = migrate_nonblock(fd, fdv);
1993         if (rc < 0) {
1994                 llapi_lease_release(fd);
1995                 goto out;
1996         }
1997
1998         rc = migrate_set_timestamps(fd, &st);
1999         if (rc < 0) {
2000                 error_loc = "cannot set source file timestamp";
2001                 goto out;
2002         }
2003
2004         /* Atomically put lease, merge layouts and close. */
2005         data = calloc(1, offsetof(typeof(*data), lil_ids[1]));
2006         if (!data) {
2007                 error_loc = "memory allocation";
2008                 goto out;
2009         }
2010         data->lil_mode = LL_LEASE_UNLCK;
2011         data->lil_flags = LL_LEASE_LAYOUT_MERGE;
2012         data->lil_count = 1;
2013         data->lil_ids[0] = fdv;
2014         rc = llapi_lease_set(fd, data);
2015         if (rc < 0) {
2016                 error_loc = "cannot merge layout";
2017                 goto out;
2018         } else if (rc == 0) {
2019                 rc = -EBUSY;
2020                 error_loc = "lost lease lock";
2021                 goto out;
2022         }
2023         rc = 0;
2024
2025 out:
2026         if (data)
2027                 free(data);
2028         if (fd >= 0)
2029                 close(fd);
2030         if (fdv >= 0)
2031                 close(fdv);
2032         if (rc < 0)
2033                 fprintf(stderr, "error: %s: %s: %s: %s\n",
2034                         progname, name, error_loc, strerror(-rc));
2035         return rc;
2036 }
2037
2038 static int mirror_extend(char *fname, struct mirror_args *mirror_list,
2039                          enum mirror_flags mirror_flags)
2040 {
2041         int rc = 0;
2042
2043         while (mirror_list) {
2044                 if (mirror_list->m_file) {
2045                         rc = mirror_extend_file(fname, mirror_list->m_file,
2046                                                 mirror_flags);
2047                 } else {
2048                         __u32 mirror_count = mirror_list->m_count;
2049
2050                         while (mirror_count > 0) {
2051                                 rc = mirror_extend_layout(fname,
2052                                                         mirror_list->m_layout,
2053                                                         mirror_list->m_inherit,
2054                                                         mirror_list->m_flags);
2055                                 if (rc)
2056                                         break;
2057
2058                                 --mirror_count;
2059                         }
2060                 }
2061                 if (rc)
2062                         break;
2063
2064                 mirror_list = mirror_list->m_next;
2065         }
2066
2067         return rc;
2068 }
2069
2070 static int find_mirror_id(struct llapi_layout *layout, void *cbdata)
2071 {
2072         uint32_t id;
2073         int rc;
2074
2075         rc = llapi_layout_mirror_id_get(layout, &id);
2076         if (rc < 0)
2077                 return rc;
2078
2079         if ((__u16)id == *(__u16 *)cbdata)
2080                 return LLAPI_LAYOUT_ITER_STOP;
2081
2082         return LLAPI_LAYOUT_ITER_CONT;
2083 }
2084
2085 static int find_comp_id(struct llapi_layout *layout, void *cbdata)
2086 {
2087         uint32_t id;
2088         int rc;
2089
2090         rc = llapi_layout_comp_id_get(layout, &id);
2091         if (rc < 0)
2092                 return rc;
2093
2094         if (id == *(__u32 *)cbdata)
2095                 return LLAPI_LAYOUT_ITER_STOP;
2096
2097         return LLAPI_LAYOUT_ITER_CONT;
2098 }
2099
2100 static int find_mirror_id_by_pool(struct llapi_layout *layout, void *cbdata)
2101 {
2102         char buf[LOV_MAXPOOLNAME + 1];
2103         struct pool_to_id_cbdata *d = (void *)cbdata;
2104         uint32_t id;
2105         int rc;
2106
2107         rc = llapi_layout_pool_name_get(layout, buf, sizeof(buf));
2108         if (rc < 0)
2109                 return rc;
2110         if (strcmp(d->pool, buf))
2111                 return LLAPI_LAYOUT_ITER_CONT;
2112
2113         rc = llapi_layout_mirror_id_get(layout, &id);
2114         if (rc < 0)
2115                 return rc;
2116         d->id = id;
2117
2118         return LLAPI_LAYOUT_ITER_STOP;
2119 }
2120
2121 static int find_comp_id_by_pool(struct llapi_layout *layout, void *cbdata)
2122 {
2123         char buf[LOV_MAXPOOLNAME + 1];
2124         struct pool_to_id_cbdata *d = (void *)cbdata;
2125         uint32_t id;
2126         int rc;
2127
2128         rc = llapi_layout_pool_name_get(layout, buf, sizeof(buf));
2129         if (rc < 0)
2130                 return rc;
2131         if (strcmp(d->pool, buf))
2132                 return LLAPI_LAYOUT_ITER_CONT;
2133
2134         rc = llapi_layout_comp_id_get(layout, &id);
2135         if (rc < 0)
2136                 return rc;
2137         d->id = id;
2138
2139         return LLAPI_LAYOUT_ITER_STOP;
2140 }
2141
2142 struct collect_ids_data {
2143         __u16   *cid_ids;
2144         int     cid_count;
2145         __u16   cid_exclude;
2146 };
2147
2148 static int collect_mirror_id(struct llapi_layout *layout, void *cbdata)
2149 {
2150         struct collect_ids_data *cid = cbdata;
2151         uint32_t id;
2152         int rc;
2153
2154         rc = llapi_layout_mirror_id_get(layout, &id);
2155         if (rc < 0)
2156                 return rc;
2157
2158         if ((__u16)id != cid->cid_exclude) {
2159                 int i;
2160
2161                 for (i = 0; i < cid->cid_count; i++) {
2162                         /* already collected the mirror id */
2163                         if (id == cid->cid_ids[i])
2164                                 return LLAPI_LAYOUT_ITER_CONT;
2165                 }
2166                 cid->cid_ids[cid->cid_count] = id;
2167                 cid->cid_count++;
2168         }
2169
2170         return LLAPI_LAYOUT_ITER_CONT;
2171 }
2172
2173 /**
2174  * last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2175  * @mirror_id: Mirror id to be checked.
2176  * @layout:    Mirror component list.
2177  *
2178  * This function checks if a mirror with specified @mirror_id is the last
2179  * non-stale mirror of a layout @layout.
2180  *
2181  * Return: true or false.
2182  */
2183 static inline
2184 bool last_non_stale_mirror(__u16 mirror_id, struct llapi_layout *layout)
2185 {
2186         __u16 mirror_ids[128] = { 0 };
2187         struct collect_ids_data cid = { .cid_ids = mirror_ids,
2188                                         .cid_count = 0,
2189                                         .cid_exclude = mirror_id, };
2190         int i;
2191
2192         llapi_layout_comp_iterate(layout, collect_mirror_id, &cid);
2193
2194         for (i = 0; i < cid.cid_count; i++) {
2195                 struct llapi_resync_comp comp_array[1024] = { { 0 } };
2196                 int comp_size = 0;
2197
2198                 comp_size = llapi_mirror_find_stale(layout, comp_array,
2199                                                     ARRAY_SIZE(comp_array),
2200                                                     &mirror_ids[i], 1);
2201                 if (comp_size == 0)
2202                         return false;
2203         }
2204
2205         return true;
2206 }
2207
2208 static int mirror_split(const char *fname, __u32 id, const char *pool,
2209                         enum mirror_flags mflags, const char *victim_file)
2210 {
2211         struct llapi_layout *layout;
2212         char parent[PATH_MAX];
2213         char victim[PATH_MAX];
2214         int flags = O_CREAT | O_EXCL | O_LOV_DELAY_CREATE | O_NOFOLLOW;
2215         char *ptr;
2216         struct ll_ioc_lease *data;
2217         uint16_t mirror_count;
2218         __u32 mirror_id;
2219         int mdt_index;
2220         int fd, fdv;
2221         bool purge = true; /* delete mirror by setting fdv=fd */
2222         bool is_encrypted;
2223         int rc;
2224
2225         if (victim_file && (strcmp(fname, victim_file) == 0)) {
2226                 fprintf(stderr,
2227                         "error %s: the source file '%s' and -f file are the same\n",
2228                         progname, fname);
2229                 return -EINVAL;
2230         }
2231
2232         /* check fname contains mirror with mirror_id/comp_id */
2233         layout = llapi_layout_get_by_path(fname, 0);
2234         if (!layout) {
2235                 fprintf(stderr,
2236                         "error %s: file '%s' couldn't get layout\n",
2237                         progname, fname);
2238                 return -EINVAL;
2239         }
2240
2241         rc = llapi_layout_sanity(layout, fname, false, true);
2242         if (rc) {
2243                 llapi_layout_sanity_perror(rc);
2244                 goto free_layout;
2245         }
2246
2247         rc = llapi_layout_mirror_count_get(layout, &mirror_count);
2248         if (rc) {
2249                 fprintf(stderr,
2250                         "error %s: file '%s' couldn't get mirror count\n",
2251                         progname, fname);
2252                 goto free_layout;
2253         }
2254         if (mirror_count < 2) {
2255                 fprintf(stderr,
2256                         "error %s: file '%s' has %d component, cannot split\n",
2257                         progname, fname, mirror_count);
2258                 goto free_layout;
2259         }
2260
2261         if (mflags & MF_COMP_POOL) {
2262                 struct pool_to_id_cbdata data = { .pool = pool };
2263
2264                 rc = llapi_layout_comp_iterate(layout, find_mirror_id_by_pool,
2265                                                &data);
2266                 mirror_id = data.id;
2267         } else if (mflags & MF_COMP_ID) {
2268                 rc = llapi_layout_comp_iterate(layout, find_comp_id, &id);
2269                 mirror_id = mirror_id_of(id);
2270         } else {
2271                 rc = llapi_layout_comp_iterate(layout, find_mirror_id, &id);
2272                 mirror_id = id;
2273         }
2274         if (rc < 0) {
2275                 fprintf(stderr, "error %s: failed to iterate layout of '%s'\n",
2276                         progname, fname);
2277                 goto free_layout;
2278         } else if (rc == LLAPI_LAYOUT_ITER_CONT) {
2279                 if (mflags & MF_COMP_POOL) {
2280                         fprintf(stderr,
2281                                 "error %s: file '%s' does not contain mirror with pool '%s'\n",
2282                                 progname, fname, pool);
2283                         goto free_layout;
2284                 } else if (mflags & MF_COMP_ID) {
2285                         fprintf(stderr,
2286                                 "error %s: file '%s' does not contain mirror with comp-id %u\n",
2287                                 progname, fname, id);
2288                         goto free_layout;
2289                 } else {
2290                         fprintf(stderr,
2291                                 "error %s: file '%s' does not contain mirror with id %u\n",
2292                                 progname, fname, id);
2293                         goto free_layout;
2294                 }
2295         }
2296
2297         fd = open(fname, O_RDWR);
2298         if (fd < 0) {
2299                 fprintf(stderr,
2300                         "error %s: open file '%s' failed: %s\n",
2301                         progname, fname, strerror(errno));
2302                 goto free_layout;
2303         }
2304
2305         /* get victim file directory pathname */
2306         if (strlen(fname) > sizeof(parent) - 1) {
2307                 fprintf(stderr, "error %s: file name of '%s' too long\n",
2308                         progname, fname);
2309                 rc = -ERANGE;
2310                 goto close_fd;
2311         }
2312         strncpy(parent, fname, sizeof(parent));
2313         ptr = strrchr(parent, '/');
2314         if (!ptr) {
2315                 if (!getcwd(parent, sizeof(parent))) {
2316                         fprintf(stderr, "error %s: getcwd failed: %s\n",
2317                                 progname, strerror(errno));
2318                         rc = -errno;
2319                         goto close_fd;
2320                 }
2321         } else {
2322                 if (ptr == parent)
2323                         ptr = parent + 1;
2324                 *ptr = '\0';
2325         }
2326
2327         rc = llapi_file_fget_mdtidx(fd, &mdt_index);
2328         if (rc < 0) {
2329                 fprintf(stderr, "%s: cannot get MDT index of '%s'\n",
2330                         progname, fname);
2331                 goto close_fd;
2332         }
2333
2334         rc = llapi_file_is_encrypted(fd);
2335         if (rc < 0) {
2336                 fprintf(stderr, "%s: cannot get flags of '%s': %d\n",
2337                         progname, fname, rc);
2338                 goto close_fd;
2339         }
2340         is_encrypted = rc;
2341
2342 again:
2343         if (!victim_file) {
2344                 /* use a temp file to store the splitted layout */
2345                 if (mflags & MF_DESTROY) {
2346                         char file_path[PATH_MAX];
2347                         unsigned int rnumber;
2348                         int open_flags;
2349
2350                         if (last_non_stale_mirror(mirror_id, layout)) {
2351                                 rc = -EUCLEAN;
2352                                 fprintf(stderr,
2353                                         "%s: cannot destroy the last non-stale mirror of file '%s'\n",
2354                                         progname, fname);
2355                                 goto close_fd;
2356                         }
2357
2358                         if (purge) {
2359                                 /* don't use volatile file for mirror destroy */
2360                                 fdv = fd;
2361                         } else {
2362                                 /**
2363                                  * try the old way to delete mirror using
2364                                  * volatile file.
2365                                  */
2366                                 do {
2367                                         rnumber = random();
2368                                         rc = snprintf(file_path,
2369                                                       sizeof(file_path),
2370                                                       "%s/" LUSTRE_VOLATILE_HDR ":%.4X:%.4X:fd=%.2d",
2371                                                       parent, mdt_index,
2372                                                       rnumber, fd);
2373                                         if (rc < 0 ||
2374                                             rc >= sizeof(file_path)) {
2375                                                 fdv = -ENAMETOOLONG;
2376                                                 break;
2377                                         }
2378
2379                                         open_flags = O_RDWR |
2380                                              (O_LOV_DELAY_CREATE & ~O_ACCMODE) |
2381                                              O_CREAT | O_EXCL | O_NOFOLLOW;
2382                                         fdv = open(file_path, open_flags,
2383                                                    S_IRUSR | S_IWUSR);
2384                                         if (fdv < 0)
2385                                                 rc = -errno;
2386                                 } while (fdv < 0 && rc == -EEXIST);
2387                         }
2388                 } else {
2389                         if (is_encrypted) {
2390                                 rc = -1;
2391                                 fprintf(stderr,
2392                                         "error %s: not permitted on encrypted file '%s': %d\n",
2393                                         progname, fname, rc);
2394                                 goto close_fd;
2395                         }
2396
2397                         snprintf(victim, sizeof(victim), "%s.mirror~%u",
2398                                  fname, mirror_id);
2399                         fdv = open(victim, flags, S_IRUSR | S_IWUSR);
2400                 }
2401         } else {
2402                 /* user specified victim file */
2403                 if (is_encrypted) {
2404                         rc = -1;
2405                         fprintf(stderr,
2406                                 "error %s: not permitted on encrypted file '%s': %d\n",
2407                                 progname, fname, rc);
2408                         goto close_fd;
2409                 }
2410                 fdv = open(victim_file, flags, S_IRUSR | S_IWUSR);
2411         }
2412
2413         if (fdv < 0) {
2414                 fprintf(stderr,
2415                         "error %s: create victim file failed: %s\n",
2416                         progname, strerror(errno));
2417                 goto close_fd;
2418         }
2419
2420         /* get lease lock of fname */
2421         rc = llapi_lease_acquire(fd, LL_LEASE_WRLCK);
2422         if (rc < 0) {
2423                 fprintf(stderr,
2424                         "error %s: cannot get lease of file '%s': %d\n",
2425                         progname, fname, rc);
2426                 goto close_victim;
2427         }
2428
2429         /* Atomatically put lease, split layouts and close. */
2430         data = malloc(offsetof(typeof(*data), lil_ids[2]));
2431         if (!data) {
2432                 rc = -ENOMEM;
2433                 goto close_victim;
2434         }
2435
2436         data->lil_mode = LL_LEASE_UNLCK;
2437         data->lil_flags = LL_LEASE_LAYOUT_SPLIT;
2438         data->lil_count = 2;
2439         data->lil_ids[0] = fdv;
2440         data->lil_ids[1] = mirror_id;
2441         rc = llapi_lease_set(fd, data);
2442         if (rc <= 0) {
2443                 if (rc == -EINVAL && purge) {
2444                         /* could be old MDS which prohibit fd==fdv */
2445                         purge = false;
2446                         goto again;
2447
2448                 }
2449                 if (rc == 0) /* lost lease lock */
2450                         rc = -EBUSY;
2451                 fprintf(stderr,
2452                         "error %s: cannot split '%s': %s\n",
2453                         progname, fname, strerror(-rc));
2454         } else {
2455                 rc = 0;
2456         }
2457         free(data);
2458
2459 close_victim:
2460         if (!purge)
2461                 close(fdv);
2462 close_fd:
2463         close(fd);
2464 free_layout:
2465         llapi_layout_free(layout);
2466         return rc;
2467 }
2468
2469 static inline
2470 int lfs_mirror_resync_file(const char *fname, struct ll_ioc_lease *ioc,
2471                            __u16 *mirror_ids, int ids_nr);
2472
2473 static int lfs_migrate_to_dom(int fd, int fdv, char *name,
2474                               __u64 migration_flags,
2475                               struct llapi_stripe_param *param,
2476                               struct llapi_layout *layout)
2477 {
2478         struct ll_ioc_lease *data = NULL;
2479         int rc;
2480
2481         rc = llapi_lease_acquire(fd, LL_LEASE_RDLCK);
2482         if (rc < 0) {
2483                 error_loc = "cannot get lease";
2484                 goto out_close;
2485         }
2486
2487         /* Atomically put lease, merge layouts, resync and close. */
2488         data = calloc(1, offsetof(typeof(*data), lil_ids[1024]));
2489         if (!data) {
2490                 error_loc = "memory allocation";
2491                 goto out_close;
2492         }
2493         data->lil_mode = LL_LEASE_UNLCK;
2494         data->lil_flags = LL_LEASE_LAYOUT_MERGE;
2495         data->lil_count = 1;
2496         data->lil_ids[0] = fdv;
2497         rc = llapi_lease_set(fd, data);
2498         if (rc < 0) {
2499                 error_loc = "cannot merge layout";
2500                 goto out_close;
2501         } else if (rc == 0) {
2502                 rc = -EBUSY;
2503                 error_loc = "lost lease lock";
2504                 goto out_close;
2505         }
2506         close(fd);
2507         close(fdv);
2508
2509         rc = lfs_mirror_resync_file(name, data, NULL, 0);
2510         if (rc) {
2511                 error_loc = "cannot resync file";
2512                 goto out;
2513         }
2514
2515         /* delete first mirror now */
2516         rc = mirror_split(name, 1, NULL, MF_DESTROY, NULL);
2517         if (rc < 0)
2518                 error_loc = "cannot delete old layout";
2519         goto out;
2520
2521 out_close:
2522         close(fd);
2523         close(fdv);
2524 out:
2525         if (rc < 0)
2526                 fprintf(stderr, "error: %s: %s: %s: %s\n",
2527                         progname, name, error_loc, strerror(-rc));
2528         else if (migration_flags & MIGRATION_VERBOSE)
2529                 printf("%s\n", name);
2530         if (data)
2531                 free(data);
2532         return rc;
2533 }
2534
2535 /**
2536  * Parse a string containing an target index list into an array of integers.
2537  *
2538  * The input string contains a comma delimited list of individual
2539  * indices and ranges, for example "1,2-4,7". Add the indices into the
2540  * \a tgts array and remove duplicates.
2541  *
2542  * \param[out] tgts             array to store indices in
2543  * \param[in] size              size of \a tgts array
2544  * \param[in] offset            starting index in \a tgts
2545  * \param[in] arg               string containing OST index list
2546  * \param[in/out] overstriping  index list may contain duplicates
2547  *
2548  * \retval positive    number of indices in \a tgts
2549  * \retval -EINVAL     unable to parse \a arg
2550  */
2551 static int parse_targets(__u32 *tgts, int size, int offset, char *arg,
2552                          unsigned long long *pattern)
2553 {
2554         int rc;
2555         int nr = offset;
2556         int slots = size - offset;
2557         char *ptr = NULL;
2558         bool overstriped = false;
2559         bool end_of_loop;
2560
2561         if (!arg)
2562                 return -EINVAL;
2563
2564         end_of_loop = false;
2565         while (!end_of_loop) {
2566                 int start_index = 0;
2567                 int end_index = 0;
2568                 int i;
2569                 char *endptr = NULL;
2570
2571                 rc = -EINVAL;
2572
2573                 ptr = strchrnul(arg, ',');
2574
2575                 end_of_loop = *ptr == '\0';
2576                 *ptr = '\0';
2577
2578                 errno = 0;
2579                 start_index = strtol(arg, &endptr, 0);
2580                 if (endptr == arg) /* no data at all */
2581                         break;
2582                 if (errno != 0 || start_index < -1 ||
2583                     (*endptr != '-' && *endptr != '\0'))
2584                         break;
2585
2586                 end_index = start_index;
2587                 if (*endptr == '-') {
2588                         errno = 0;
2589                         end_index = strtol(endptr + 1, &endptr, 0);
2590                         if (errno != 0 || *endptr != '\0' || end_index < -1)
2591                                 break;
2592                         if (end_index < start_index)
2593                                 break;
2594                 }
2595
2596                 for (i = start_index; i <= end_index && slots > 0; i++) {
2597                         int j;
2598
2599                         /* remove duplicate */
2600                         for (j = 0; j < offset; j++) {
2601                                 if (tgts[j] == i && pattern &&
2602                                     *pattern == LLAPI_LAYOUT_OVERSTRIPING)
2603                                         overstriped = true;
2604                                 else if (tgts[j] == i)
2605                                         return -EINVAL;
2606                         }
2607
2608                         j = offset;
2609
2610                         if (j == offset) { /* check complete */
2611                                 tgts[nr++] = i;
2612                                 --slots;
2613                         }
2614                 }
2615
2616                 if (slots == 0 && i < end_index)
2617                         break;
2618
2619                 *ptr = ',';
2620                 arg = ++ptr;
2621                 offset = nr;
2622                 rc = 0;
2623         }
2624         if (!end_of_loop && ptr)
2625                 *ptr = ',';
2626
2627         if (!overstriped && pattern)
2628                 *pattern = LLAPI_LAYOUT_DEFAULT;
2629
2630         return rc < 0 ? rc : nr;
2631 }
2632
2633 struct lfs_setstripe_args {
2634         unsigned long long       lsa_comp_end;
2635         unsigned long long       lsa_stripe_size;
2636         unsigned long long       lsa_extension_size;
2637         long long                lsa_stripe_count;
2638         long long                lsa_stripe_off;
2639         __u32                    lsa_comp_flags;
2640         __u32                    lsa_comp_neg_flags;
2641         unsigned long long       lsa_pattern;
2642         unsigned int             lsa_mirror_count;
2643         int                      lsa_nr_tgts;
2644         bool                     lsa_first_comp;
2645         bool                     lsa_extension_comp;
2646         __u32                   *lsa_tgts;
2647         char                    *lsa_pool_name;
2648 };
2649
2650 static inline void setstripe_args_init(struct lfs_setstripe_args *lsa)
2651 {
2652         unsigned int mirror_count = lsa->lsa_mirror_count;
2653         bool first_comp = lsa->lsa_first_comp;
2654
2655         memset(lsa, 0, sizeof(*lsa));
2656
2657         lsa->lsa_stripe_size = LLAPI_LAYOUT_DEFAULT;
2658         lsa->lsa_stripe_count = LLAPI_LAYOUT_DEFAULT;
2659         lsa->lsa_stripe_off = LLAPI_LAYOUT_DEFAULT;
2660         lsa->lsa_pattern = LLAPI_LAYOUT_RAID0;
2661         lsa->lsa_pool_name = NULL;
2662
2663         lsa->lsa_mirror_count = mirror_count;
2664         lsa->lsa_first_comp = first_comp;
2665 }
2666
2667 /**
2668  * setstripe_args_init_inherit() - Initialize and inherit stripe options.
2669  * @lsa: Stripe options to be initialized and inherited.
2670  *
2671  * This function initializes stripe options in @lsa and inherit
2672  * stripe_size, stripe_count and OST pool_name options.
2673  *
2674  * Return: void.
2675  */
2676 static inline void setstripe_args_init_inherit(struct lfs_setstripe_args *lsa)
2677 {
2678         unsigned long long stripe_size;
2679         long long stripe_count;
2680         char *pool_name = NULL;
2681
2682         stripe_size = lsa->lsa_stripe_size;
2683         stripe_count = lsa->lsa_stripe_count;
2684         pool_name = lsa->lsa_pool_name;
2685
2686         setstripe_args_init(lsa);
2687
2688         lsa->lsa_stripe_size = stripe_size;
2689         lsa->lsa_stripe_count = stripe_count;
2690         lsa->lsa_pool_name = pool_name;
2691 }
2692
2693 static inline bool setstripe_args_specified(struct lfs_setstripe_args *lsa)
2694 {
2695         return (lsa->lsa_stripe_size != LLAPI_LAYOUT_DEFAULT ||
2696                 lsa->lsa_stripe_count != LLAPI_LAYOUT_DEFAULT ||
2697                 lsa->lsa_stripe_off != LLAPI_LAYOUT_DEFAULT ||
2698                 lsa->lsa_pattern != LLAPI_LAYOUT_RAID0 ||
2699                 lsa->lsa_comp_end != 0);
2700 }
2701
2702 static int lsa_args_stripe_count_check(struct lfs_setstripe_args *lsa)
2703 {
2704         if (lsa->lsa_nr_tgts) {
2705                 if (lsa->lsa_nr_tgts < 0 ||
2706                     lsa->lsa_nr_tgts >= LOV_MAX_STRIPE_COUNT) {
2707                         fprintf(stderr, "Invalid nr_tgts(%d)\n",
2708                                 lsa->lsa_nr_tgts);
2709                         errno = EINVAL;
2710                         return -1;
2711                 }
2712
2713                 if (lsa->lsa_stripe_count > 0 &&
2714                     lsa->lsa_stripe_count != LLAPI_LAYOUT_DEFAULT &&
2715                     lsa->lsa_stripe_count != LLAPI_LAYOUT_WIDE &&
2716                     lsa->lsa_nr_tgts != lsa->lsa_stripe_count) {
2717                         fprintf(stderr, "stripe_count(%lld) != nr_tgts(%d)\n",
2718                                 lsa->lsa_stripe_count,
2719                                 lsa->lsa_nr_tgts);
2720                         errno = EINVAL;
2721                         return -1;
2722                 }
2723         }
2724
2725         return 0;
2726
2727 }
2728
2729 /**
2730  * comp_args_to_layout() - Create or extend a composite layout.
2731  * @composite:       Pointer to the composite layout.
2732  * @lsa:             Stripe options for the new component.
2733  *
2734  * This function creates or extends a composite layout by adding a new
2735  * component with stripe options from @lsa.
2736  *
2737  * When modified, adjust llapi_stripe_param_verify() if needed as well.
2738  *
2739  * Return: 0 on success or an error code on failure.
2740  */
2741 static int comp_args_to_layout(struct llapi_layout **composite,
2742                                struct lfs_setstripe_args *lsa,
2743                                bool set_extent)
2744 {
2745         struct llapi_layout *layout = *composite;
2746         uint64_t prev_end = 0;
2747         uint64_t size;
2748         int i = 0, rc;
2749
2750 new_comp:
2751         if (!layout) {
2752                 layout = llapi_layout_alloc();
2753                 if (!layout) {
2754                         fprintf(stderr, "Alloc llapi_layout failed. %s\n",
2755                                 strerror(errno));
2756                         errno = ENOMEM;
2757                         return -1;
2758                 }
2759                 *composite = layout;
2760                 lsa->lsa_first_comp = true;
2761         } else {
2762                 uint64_t start;
2763
2764                 /*
2765                  * Get current component extent, current component
2766                  * must be the tail component.
2767                  */
2768                 rc = llapi_layout_comp_extent_get(layout, &start, &prev_end);
2769                 if (rc) {
2770                         fprintf(stderr, "Get comp extent failed. %s\n",
2771                                 strerror(errno));
2772                         return rc;
2773                 }
2774
2775                 if (lsa->lsa_first_comp) {
2776                         prev_end = 0;
2777                         rc = llapi_layout_add_first_comp(layout);
2778                 } else {
2779                         rc = llapi_layout_comp_add(layout);
2780                 }
2781                 if (rc) {
2782                         fprintf(stderr, "Add component failed. %s\n",
2783                                 strerror(errno));
2784                         return rc;
2785                 }
2786         }
2787
2788         rc = llapi_layout_comp_flags_set(layout, lsa->lsa_comp_flags);
2789         if (rc) {
2790                 fprintf(stderr, "Set flags 0x%x failed: %s\n",
2791                         lsa->lsa_comp_flags, strerror(errno));
2792                 return rc;
2793         }
2794
2795         if (set_extent) {
2796                 uint64_t comp_end = lsa->lsa_comp_end;
2797
2798                 /*
2799                  * The extendable component is 0-length, so it can be removed
2800                  * if there is insufficient space to extend it.
2801                  */
2802                 if (lsa->lsa_extension_comp)
2803                         comp_end = prev_end;
2804
2805                 rc = llapi_layout_comp_extent_set(layout, prev_end,
2806                                                   comp_end);
2807                 if (rc) {
2808                         fprintf(stderr, "Set extent [%lu, %lu) failed. %s\n",
2809                                 prev_end, comp_end, strerror(errno));
2810                         return rc;
2811                 }
2812         }
2813         /* reset lsa_first_comp */
2814         lsa->lsa_first_comp = false;
2815
2816         /* Data-on-MDT component setting */
2817         if (lsa->lsa_pattern == LLAPI_LAYOUT_MDT) {
2818                 /* Yaml support */
2819                 if (lsa->lsa_stripe_count == 0)
2820                         lsa->lsa_stripe_count = LLAPI_LAYOUT_DEFAULT;
2821                 if (lsa->lsa_stripe_size == lsa->lsa_comp_end)
2822                         lsa->lsa_stripe_size = LLAPI_LAYOUT_DEFAULT;
2823                 if (lsa->lsa_stripe_off == -1 ||
2824                     lsa->lsa_stripe_off == 0)
2825                         lsa->lsa_stripe_off = LLAPI_LAYOUT_DEFAULT;
2826                 /*
2827                  * In case of Data-on-MDT patterns the only extra option
2828                  * applicable is stripe size option.
2829                  */
2830                 if (lsa->lsa_stripe_count != LLAPI_LAYOUT_DEFAULT) {
2831                         fprintf(stderr,
2832                                 "Option 'stripe-count' can't be specified with Data-on-MDT component: %lld\n",
2833                                 lsa->lsa_stripe_count);
2834                         errno = EINVAL;
2835                         return -1;
2836                 }
2837                 if (lsa->lsa_stripe_size != LLAPI_LAYOUT_DEFAULT &&
2838                     lsa->lsa_stripe_size != lsa->lsa_comp_end - prev_end) {
2839                         fprintf(stderr,
2840                                 "Option 'stripe-size' can't be specified with Data-on-MDT component: %llu\n",
2841                                 lsa->lsa_stripe_size);
2842                         errno = EINVAL;
2843                         return -1;
2844                 }
2845                 if (lsa->lsa_nr_tgts != 0) {
2846                         fprintf(stderr,
2847                                 "Option 'ost-list' can't be specified with Data-on-MDT component: '%i'\n",
2848                                 lsa->lsa_nr_tgts);
2849                         errno = EINVAL;
2850                         return -1;
2851                 }
2852                 if (lsa->lsa_stripe_off != LLAPI_LAYOUT_DEFAULT) {
2853                         fprintf(stderr,
2854                                 "Option 'stripe-offset' can't be specified with Data-on-MDT component: %lld\n",
2855                                 lsa->lsa_stripe_off);
2856                         errno = EINVAL;
2857                         return -1;
2858                 }
2859                 if (lsa->lsa_pool_name != 0) {
2860                         fprintf(stderr,
2861                                 "Option 'pool' can't be specified with Data-on-MDT component: '%s'\n",
2862                                 lsa->lsa_pool_name);
2863                         errno = EINVAL;
2864                         return -1;
2865                 }
2866
2867                 rc = llapi_layout_pattern_set(layout, lsa->lsa_pattern);
2868                 if (rc) {
2869                         fprintf(stderr, "Set stripe pattern %#llx failed. %s\n",
2870                                 lsa->lsa_pattern,
2871                                 strerror(errno));
2872                         return rc;
2873                 }
2874                 /* Data-on-MDT component has always single stripe up to end */
2875                 lsa->lsa_stripe_size = lsa->lsa_comp_end;
2876         } else if (lsa->lsa_pattern == LLAPI_LAYOUT_OVERSTRIPING) {
2877                 rc = llapi_layout_pattern_set(layout, lsa->lsa_pattern);
2878                 if (rc) {
2879                         fprintf(stderr, "Set stripe pattern %#llx failed. %s\n",
2880                                 lsa->lsa_pattern,
2881                                 strerror(errno));
2882                         return rc;
2883                 }
2884         }
2885
2886         size = lsa->lsa_comp_flags & LCME_FL_EXTENSION ?
2887                 lsa->lsa_extension_size : lsa->lsa_stripe_size;
2888
2889         if (lsa->lsa_comp_flags & LCME_FL_EXTENSION)
2890                 rc = llapi_layout_extension_size_set(layout, size);
2891         else
2892                 rc = llapi_layout_stripe_size_set(layout, size);
2893
2894         if (rc) {
2895                 fprintf(stderr, "Set stripe size %lu failed: %s\n",
2896                         size, strerror(errno));
2897                 return rc;
2898         }
2899
2900         rc = llapi_layout_stripe_count_set(layout, lsa->lsa_stripe_count);
2901         if (rc) {
2902                 fprintf(stderr, "Set stripe count %lld failed: %s\n",
2903                         lsa->lsa_stripe_count, strerror(errno));
2904                 return rc;
2905         }
2906
2907         if (lsa->lsa_pool_name) {
2908                 rc = llapi_layout_pool_name_set(layout, lsa->lsa_pool_name);
2909                 if (rc) {
2910                         fprintf(stderr, "Set pool name: %s failed. %s\n",
2911                                 lsa->lsa_pool_name, strerror(errno));
2912                         return rc;
2913                 }
2914         } else {
2915                 rc = llapi_layout_pool_name_set(layout, "");
2916                 if (rc) {
2917                         fprintf(stderr, "Clear pool name failed: %s\n",
2918                                 strerror(errno));
2919                         return rc;
2920                 }
2921         }
2922
2923         rc = lsa_args_stripe_count_check(lsa);
2924         if (rc)
2925                 return rc;
2926
2927         if (lsa->lsa_nr_tgts > 0) {
2928                 bool found = false;
2929
2930                 for (i = 0; i < lsa->lsa_nr_tgts; i++) {
2931                         rc = llapi_layout_ost_index_set(layout, i,
2932                                                         lsa->lsa_tgts[i]);
2933                         if (rc)
2934                                 break;
2935
2936                         /* Make sure stripe offset is in OST list. */
2937                         if (lsa->lsa_tgts[i] == lsa->lsa_stripe_off)
2938                                 found = true;
2939                 }
2940                 if (!found) {
2941                         fprintf(stderr, "Invalid stripe offset '%lld', not in the target list",
2942                                 lsa->lsa_stripe_off);
2943                         errno = EINVAL;
2944                         return -1;
2945                 }
2946         } else if (lsa->lsa_stripe_off != LLAPI_LAYOUT_DEFAULT &&
2947                    lsa->lsa_stripe_off != -1) {
2948                 rc = llapi_layout_ost_index_set(layout, 0, lsa->lsa_stripe_off);
2949         }
2950         if (rc) {
2951                 fprintf(stderr, "Set ost index %d failed. %s\n",
2952                         i, strerror(errno));
2953                 return rc;
2954         }
2955
2956         /* Create the second, virtual component of extension space */
2957         if (lsa->lsa_extension_comp) {
2958                 lsa->lsa_comp_flags |= LCME_FL_EXTENSION;
2959                 lsa->lsa_extension_comp = false;
2960                 goto new_comp;
2961         }
2962
2963         return rc;
2964 }
2965
2966 static int build_component(struct llapi_layout **layout,
2967                            struct lfs_setstripe_args *lsa, bool set_extent)
2968 {
2969         int rc;
2970
2971         rc = comp_args_to_layout(layout, lsa, set_extent);
2972         if (rc)
2973                 return rc;
2974
2975         if (lsa->lsa_mirror_count > 0) {
2976                 rc = llapi_layout_mirror_count_set(*layout,
2977                                                    lsa->lsa_mirror_count);
2978                 if (rc)
2979                         return rc;
2980
2981                 rc = llapi_layout_flags_set(*layout, LCM_FL_RDONLY);
2982                 if (rc)
2983                         return rc;
2984                 lsa->lsa_mirror_count = 0;
2985         }
2986
2987         return rc;
2988 }
2989
2990 static int build_prev_component(struct llapi_layout **layout,
2991                                 struct lfs_setstripe_args *prev,
2992                                 struct lfs_setstripe_args *lsa,
2993                                 bool set_extent)
2994 {
2995         int extension = lsa->lsa_comp_flags & LCME_FL_EXTENSION;
2996         int rc;
2997
2998         if (prev->lsa_stripe_size) {
2999                 if (extension) {
3000                         prev->lsa_comp_end = lsa->lsa_comp_end;
3001                         prev->lsa_extension_size = lsa->lsa_extension_size;
3002                         prev->lsa_extension_comp = true;
3003                 }
3004
3005                 rc = build_component(layout, prev, true);
3006                 if (rc)
3007                         return rc;
3008         }
3009
3010         /*
3011          * Copy lsa to previous lsa;
3012          * if this is an extension component, make the previous invalid;
3013          */
3014         if (extension)
3015                 prev->lsa_stripe_size = 0;
3016         else
3017                 *prev = *lsa;
3018
3019         return 0;
3020 }
3021
3022 static int build_layout_from_yaml_node(struct cYAML *node,
3023                                        struct llapi_layout **layout,
3024                                        struct lfs_setstripe_args *lsa,
3025                                        struct lfs_setstripe_args *prevp)
3026 {
3027         struct lfs_setstripe_args prev = { 0 };
3028         __u32 *osts = lsa->lsa_tgts;
3029         char *string;
3030         int rc = 0;
3031
3032         if (!prevp)
3033                 prevp = &prev;
3034
3035         while (node) {
3036                 string = node->cy_string;
3037
3038                 if (node->cy_type == CYAML_TYPE_OBJECT) {
3039                         /* go deep to sub blocks */
3040                         if (string && !strncmp(string, "component", 9) &&
3041                             strncmp(string, "component0", 10) &&
3042                             strncmp(string, "components", 10)) {
3043                                 rc = build_prev_component(layout, prevp, lsa,
3044                                                           true);
3045                                 if (rc)
3046                                         return rc;
3047
3048                                 /* initialize lsa. */
3049                                 setstripe_args_init(lsa);
3050                                 lsa->lsa_first_comp = false;
3051                                 lsa->lsa_tgts = osts;
3052                         }
3053
3054                         rc = build_layout_from_yaml_node(node->cy_child, layout,
3055                                                          lsa, prevp);
3056                         if (rc)
3057                                 return rc;
3058                 } else {
3059                         if (!node->cy_string)
3060                                 return -EINVAL;
3061
3062                         /* skip leading lmm_ if present, to simplify parsing */
3063                         if (strncmp(string, "lmm_", 4) == 0)
3064                                 string += 4;
3065
3066                         if (node->cy_type == CYAML_TYPE_STRING) {
3067                                 if (!strcmp(string, "lcme_extent.e_end")) {
3068                                         if (!strcmp(node->cy_valuestring, "EOF") ||
3069                                             !strcmp(node->cy_valuestring, "eof"))
3070                                                 lsa->lsa_comp_end = LUSTRE_EOF;
3071                                 } else if (!strcmp(string, "pool")) {
3072                                         lsa->lsa_pool_name = node->cy_valuestring;
3073                                 } else if (!strcmp(string, "pattern")) {
3074                                         if (!strcmp(node->cy_valuestring, "mdt"))
3075                                                 lsa->lsa_pattern = LLAPI_LAYOUT_MDT;
3076                                         if (!strcmp(node->cy_valuestring,
3077                                                     "raid0,overstriped"))
3078                                                 lsa->lsa_pattern =
3079                                                         LLAPI_LAYOUT_OVERSTRIPING;
3080                                 } else if (!strcmp(string, "lcme_flags")) {
3081                                         rc = comp_str2flags(node->cy_valuestring,
3082                                                             &lsa->lsa_comp_flags,
3083                                                             &lsa->lsa_comp_neg_flags);
3084                                         if (rc)
3085                                                 return rc;
3086                                         /*
3087                                          * Only template flags have meaning in
3088                                          * the layout for a new file
3089                                          */
3090                                         lsa->lsa_comp_flags &= LCME_TEMPLATE_FLAGS;
3091                                 }
3092                         } else if (node->cy_type == CYAML_TYPE_NUMBER) {
3093                                 if (!strcmp(string, "lcm_mirror_count")) {
3094                                         lsa->lsa_mirror_count = node->cy_valueint;
3095                                 } else if (!strcmp(string, "lcme_extent.e_start")) {
3096                                         if (node->cy_valueint == 0)
3097                                                 lsa->lsa_first_comp = true;
3098                                 } else if (!strcmp(string, "lcme_extent.e_end")) {
3099                                         if (node->cy_valueint == -1)
3100                                                 lsa->lsa_comp_end = LUSTRE_EOF;
3101                                         else
3102                                                 lsa->lsa_comp_end = node->cy_valueint;
3103                                 } else if (!strcmp(string, "stripe_count")) {
3104                                         lsa->lsa_stripe_count = node->cy_valueint;
3105                                 } else if (!strcmp(string, "stripe_size")) {
3106                                         lsa->lsa_stripe_size = node->cy_valueint;
3107                                 } else if (!strcmp(string, "extension_size")) {
3108                                         lsa->lsa_extension_size = node->cy_valueint;
3109                                         lsa->lsa_extension_comp = true;
3110                                 } else if (!strcmp(string, "stripe_offset")) {
3111                                         lsa->lsa_stripe_off = node->cy_valueint;
3112                                 } else if (!strcmp(string, "l_ost_idx")) {
3113                                         osts[lsa->lsa_nr_tgts] = node->cy_valueint;
3114                                         lsa->lsa_nr_tgts++;
3115                                 }
3116                         }
3117                 }
3118                 node = node->cy_next;
3119         }
3120
3121         if (prevp == &prev) {
3122                 rc = build_prev_component(layout, prevp, lsa, true);
3123                 if (rc)
3124                         return rc;
3125
3126                 if (!(lsa->lsa_comp_flags & LCME_FL_EXTENSION))
3127                         rc = build_component(layout, lsa, *layout != NULL);
3128         }
3129
3130         return rc;
3131 }
3132
3133 static int lfs_comp_create_from_yaml(char *template,
3134                                      struct llapi_layout **layout,
3135                                      struct lfs_setstripe_args *lsa,
3136                                      __u32 *osts)
3137 {
3138         struct cYAML *tree = NULL, *err_rc = NULL;
3139         int rc = 0;
3140
3141         tree = cYAML_build_tree(template, NULL, 0, &err_rc, false);
3142         if (!tree) {
3143                 fprintf(stderr, "%s: cannot parse YAML file %s\n",
3144                         progname, template);
3145                 cYAML_build_error(-EINVAL, -1, "yaml", "from comp yaml",
3146                                   "can't parse", &err_rc);
3147                 cYAML_print_tree2file(stderr, err_rc);
3148                 cYAML_free_tree(err_rc);
3149                 rc = -EINVAL;
3150                 goto err;
3151         }
3152
3153         /* initialize lsa for plain file */
3154         setstripe_args_init(lsa);
3155         lsa->lsa_tgts = osts;
3156
3157         rc = build_layout_from_yaml_node(tree, layout, lsa, NULL);
3158         if (rc) {
3159                 fprintf(stderr, "%s: cannot build layout from YAML file %s.\n",
3160                         progname, template);
3161                 goto err;
3162         }
3163         /* clean clean lsa */
3164         setstripe_args_init(lsa);
3165
3166 err:
3167         if (tree)
3168                 cYAML_free_tree(tree);
3169         return rc;
3170 }
3171
3172 /**
3173  * Get the extension size from the next (SEL) component and extend the
3174  * current component on it. The start of the next component is to be
3175  * adjusted as well.
3176  *
3177  * \param[in] layout    the current layout
3178  * \param[in] start     the start of the current component
3179  * \param[in,out] end   the end of the current component
3180  * \param[in] offset    the offset to adjust the end position to instead of
3181  *                      extension size
3182  *
3183  * \retval 0            - extended successfully
3184  * \retval < 0          - error
3185  */
3186 static int layout_extend_comp(struct llapi_layout *layout,
3187                               uint64_t start, uint64_t *end,
3188                               uint64_t offset)
3189 {
3190         uint64_t size, next_start, next_end;
3191         int rc;
3192
3193         rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_NEXT);
3194         if (rc < 0) {
3195                 fprintf(stderr,
3196                         "%s setstripe: cannot move component cursor: %s\n",
3197                         progname, strerror(errno));
3198                 return rc;
3199         }
3200
3201         /*
3202          * Even if the @size will not be used below, this will fail if
3203          * this is not a SEL component - a good confirmation we are
3204          * working on right components.
3205          */
3206         rc = llapi_layout_extension_size_get(layout, &size);
3207         if (rc < 0) {
3208                 fprintf(stderr,
3209                         "%s setstripe: cannot get component ext size: %s\n",
3210                         progname, strerror(errno));
3211                 return rc;
3212         }
3213
3214         rc = llapi_layout_comp_extent_get(layout, &next_start, &next_end);
3215         if (rc) {
3216                 fprintf(stderr, "%s setstripe: cannot get extent: %s\n",
3217                         progname, strerror(errno));
3218                 return rc;
3219         }
3220
3221         next_start += offset ?: size;
3222         rc = llapi_layout_comp_extent_set(layout, next_start, next_end);
3223         if (rc) {
3224                 fprintf(stderr, "%s setstripe: cannot set extent: %s\n",
3225                         progname, strerror(errno));
3226                 return rc;
3227         }
3228
3229         rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_PREV);
3230         if (rc < 0) {
3231                 fprintf(stderr,
3232                         "%s setstripe: cannot move component cursor: %s\n",
3233                         progname, strerror(errno));
3234                 return rc;
3235         }
3236
3237         *end += offset ?: size;
3238         rc = llapi_layout_comp_extent_set(layout, start, *end);
3239         if (rc) {
3240                 fprintf(stderr, "%s setstripe: cannot set extent: %s\n",
3241                         progname, strerror(errno));
3242                 return rc;
3243         }
3244
3245         return 0;
3246 }
3247
3248 /**
3249  * In 'lfs setstripe --component-add' mode, we need to fetch the extent
3250  * end of the last component in the existing file, and adjust the
3251  * first extent start of the components to be added accordingly.
3252  *
3253  * In the create mode, we need to check if the first component is an extendable
3254  * SEL component and extend its length to the extension size (first component
3255  * of the PFL file is initialised at the create time, cannot be 0-lenght.
3256  */
3257 static int layout_adjust_first_extent(char *fname, struct llapi_layout *layout,
3258                                       bool comp_add)
3259 {
3260         struct llapi_layout *head;
3261         uint64_t start = 0, prev_end = 0;
3262         uint64_t end;
3263         int rc, ret = 0;
3264
3265         if (!layout || !(comp_add || llapi_layout_is_composite(layout)))
3266                 return 0;
3267
3268         errno = 0;
3269         while (comp_add) {
3270                 head = llapi_layout_get_by_path(fname, 0);
3271                 if (!head) {
3272                         fprintf(stderr,
3273                                 "%s setstripe: cannot read layout from '%s': %s\n",
3274                                 progname, fname, strerror(errno));
3275                         return -EINVAL;
3276                 } else if (errno == ENODATA) {
3277                         /*
3278                          * file without LOVEA, this component-add will be turned
3279                          * into a component-create.
3280                          */
3281                         llapi_layout_free(head);
3282                         ret = -ENODATA;
3283
3284                         /*
3285                          * the new layout will be added to an empty one, it
3286                          * still needs to be adjusted below
3287                          */
3288                         comp_add = 0;
3289                         break;
3290                 } else if (!llapi_layout_is_composite(head)) {
3291                         fprintf(stderr,
3292                                 "%s setstripe: '%s' not a composite file\n",
3293                                 progname, fname);
3294                         llapi_layout_free(head);
3295                         return -EINVAL;
3296                 }
3297
3298                 rc = llapi_layout_comp_extent_get(head, &start, &prev_end);
3299                 if (rc) {
3300                         fprintf(stderr,
3301                                 "%s setstripe: cannot get prev extent: %s\n",
3302                                 progname, strerror(errno));
3303                         llapi_layout_free(head);
3304                         return rc;
3305                 }
3306
3307                 llapi_layout_free(head);
3308                 break;
3309         }
3310
3311         /* Make sure we use the first component of the layout to be added. */
3312         rc = llapi_layout_comp_use(layout, LLAPI_LAYOUT_COMP_USE_FIRST);
3313         if (rc < 0) {
3314                 fprintf(stderr,
3315                         "%s setstripe: cannot move component cursor: %s\n",
3316                         progname, strerror(errno));
3317                 return rc;
3318         }
3319
3320         rc = llapi_layout_comp_extent_get(layout, &start, &end);
3321         if (rc) {
3322                 fprintf(stderr, "%s setstripe: cannot get extent: %s\n",
3323                         progname, strerror(errno));
3324                 return rc;
3325         }
3326
3327         if (start == 0 && end == 0) {
3328                 rc = layout_extend_comp(layout, start, &end,
3329                                         comp_add ? prev_end : 0);
3330                 if (rc)
3331                         return rc;
3332         }
3333
3334         if (start > prev_end || end < prev_end) {
3335                 fprintf(stderr,
3336                         "%s setstripe: first extent [%lu, %lu) not adjacent with extent end %lu\n",
3337                         progname, start, end, prev_end);
3338                 return -EINVAL;
3339         }
3340
3341         rc = llapi_layout_comp_extent_set(layout, prev_end, end);
3342         if (rc) {
3343                 fprintf(stderr,
3344                         "%s setstripe: cannot set component extent [%lu, %lu): %s\n",
3345                         progname, prev_end, end, strerror(errno));
3346                 return rc;
3347         }
3348
3349         return ret;
3350 }
3351
3352 static int mirror_adjust_first_extents(struct mirror_args *list)
3353 {
3354         int rc = 0;
3355
3356         if (!list)
3357                 return 0;
3358
3359         while (list) {
3360                 rc = layout_adjust_first_extent(NULL, list->m_layout, false);
3361                 if (rc)
3362                         break;
3363                 list = list->m_next;
3364         }
3365
3366         return rc;
3367 }
3368
3369 static inline bool arg_is_eof(char *arg)
3370 {
3371         return !strncmp(arg, "-1", strlen("-1")) ||
3372                !strncmp(arg, "EOF", strlen("EOF")) ||
3373                !strncmp(arg, "eof", strlen("eof"));
3374 }
3375
3376 /**
3377  * lfs_mirror_alloc() - Allocate a mirror argument structure.
3378  *
3379  * Return: Valid mirror_args pointer on success and
3380  *         NULL if memory allocation fails.
3381  */
3382 static struct mirror_args *lfs_mirror_alloc(void)
3383 {
3384         struct mirror_args *mirror = NULL;
3385
3386         while (1) {
3387                 mirror = calloc(1, sizeof(*mirror));
3388                 if (mirror) {
3389                         mirror->m_inherit = false;
3390                         break;
3391                 }
3392
3393                 sleep(1);
3394         }
3395
3396         return mirror;
3397 }
3398
3399 /**
3400  * lfs_mirror_free() - Free memory allocated for a mirror argument
3401  *                     structure.
3402  * @mirror: Previously allocated mirror argument structure by
3403  *          lfs_mirror_alloc().
3404  *
3405  * Free memory allocated for @mirror.
3406  *
3407  * Return: void.
3408  */
3409 static void lfs_mirror_free(struct mirror_args *mirror)
3410 {
3411         if (mirror->m_layout)
3412                 llapi_layout_free(mirror->m_layout);
3413         free(mirror);
3414 }
3415
3416 /**
3417  * lfs_mirror_list_free() - Free memory allocated for a mirror list.
3418  * @mirror_list: Previously allocated mirror list.
3419  *
3420  * Free memory allocated for @mirror_list.
3421  *
3422  * Return: void.
3423  */
3424 static void lfs_mirror_list_free(struct mirror_args *mirror_list)
3425 {
3426         struct mirror_args *next_mirror = NULL;
3427
3428         while (mirror_list) {
3429                 next_mirror = mirror_list->m_next;
3430                 lfs_mirror_free(mirror_list);
3431                 mirror_list = next_mirror;
3432         }
3433 }
3434
3435 enum {
3436         LFS_POOL_OPT = 3,
3437         LFS_COMP_COUNT_OPT,
3438         LFS_COMP_START_OPT,
3439         LFS_COMP_FLAGS_OPT,
3440         LFS_COMP_DEL_OPT,
3441         LFS_COMP_SET_OPT,
3442         LFS_COMP_ADD_OPT,
3443         LFS_COMP_NO_VERIFY_OPT,
3444         LFS_PROJID_OPT,
3445         LFS_LAYOUT_FLAGS_OPT, /* used for mirror and foreign flags */
3446         LFS_MIRROR_ID_OPT,
3447         LFS_MIRROR_STATE_OPT,
3448         LFS_LAYOUT_COPY,
3449         LFS_MIRROR_INDEX_OPT,
3450         LFS_LAYOUT_FOREIGN_OPT,
3451         LFS_MODE_OPT,
3452         LFS_NEWERXY_OPT,
3453         LFS_INHERIT_RR_OPT,
3454         LFS_FIND_PERM,
3455 };
3456
3457 /* functions */
3458 static int lfs_setstripe_internal(int argc, char **argv,
3459                                   enum setstripe_origin opc)
3460 {
3461         struct lfs_setstripe_args        lsa = { 0 };
3462         struct llapi_stripe_param       *param = NULL;
3463         struct find_param                migrate_mdt_param = {
3464                 .fp_max_depth = -1,
3465                 .fp_mdt_index = -1,
3466         };
3467         char                            *fname;
3468         int                              result = 0;
3469         int                              result2 = 0;
3470         char                            *end;
3471         int                              c;
3472         int                              delete = 0;
3473         unsigned long long               size_units = 1;
3474         bool                             migrate_mode = false;
3475         bool                             migrate_mdt_mode = false;
3476         bool                             setstripe_mode = false;
3477         bool                             migration_block = false;
3478         __u64                            migration_flags = 0;
3479         __u32                            tgts[LOV_MAX_STRIPE_COUNT] = { 0 };
3480         int                              comp_del = 0, comp_set = 0;
3481         int                              comp_add = 0;
3482         __u32                            comp_id = 0;
3483         struct llapi_layout             *layout = NULL;
3484         struct llapi_layout             **lpp = &layout;
3485         bool                             mirror_mode = false;
3486         bool                             has_m_file = false;
3487         __u32                            mirror_count = 0;
3488         enum mirror_flags                mirror_flags = 0;
3489         struct mirror_args              *mirror_list = NULL;
3490         struct mirror_args              *new_mirror = NULL;
3491         struct mirror_args              *last_mirror = NULL;
3492         __u16                            mirror_id = 0;
3493         char                             cmd[PATH_MAX];
3494         bool from_yaml = false;
3495         bool from_copy = false;
3496         char *template = NULL;
3497         bool foreign_mode = false;
3498         char *xattr = NULL;
3499         uint32_t type = LU_FOREIGN_TYPE_NONE, flags = 0;
3500         char *mode_opt = NULL;
3501         mode_t previous_umask = 0;
3502         mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH;
3503
3504         struct option long_opts[] = {
3505 /* find { .val = '0',   .name = "null",         .has_arg = no_argument }, */
3506 /* find { .val = 'A',   .name = "atime",        .has_arg = required_argument }*/
3507         /* --block is only valid in migrate mode */
3508         { .val = 'b',   .name = "block",        .has_arg = no_argument },
3509 /* find { .val = 'B',   .name = "btime",        .has_arg = required_argument }*/
3510         { .val = LFS_COMP_ADD_OPT,
3511                         .name = "comp-add",     .has_arg = no_argument },
3512         { .val = LFS_COMP_ADD_OPT,
3513                         .name = "component-add", .has_arg = no_argument },
3514         { .val = LFS_COMP_DEL_OPT,
3515                         .name = "comp-del",     .has_arg = no_argument },
3516         { .val = LFS_COMP_DEL_OPT,
3517                         .name = "component-del", .has_arg = no_argument },
3518         { .val = LFS_COMP_FLAGS_OPT,
3519                         .name = "comp-flags",   .has_arg = required_argument },
3520         { .val = LFS_COMP_FLAGS_OPT,
3521                         .name = "component-flags",
3522                                                 .has_arg = required_argument },
3523         { .val = LFS_COMP_SET_OPT,
3524                         .name = "comp-set",     .has_arg = no_argument },
3525         { .val = LFS_COMP_SET_OPT,
3526                         .name = "component-set",
3527                                                 .has_arg = no_argument},
3528         { .val = LFS_COMP_NO_VERIFY_OPT,
3529                         .name = "no-verify",    .has_arg = no_argument},
3530         { .val = LFS_LAYOUT_FLAGS_OPT,
3531                         .name = "flags",        .has_arg = required_argument},
3532         { .val = LFS_LAYOUT_FOREIGN_OPT,
3533                         .name = "foreign",      .has_arg = optional_argument},
3534         { .val = LFS_MIRROR_ID_OPT,
3535                         .name = "mirror-id",    .has_arg = required_argument},
3536         { .val = LFS_MODE_OPT,
3537                         .name = "mode",         .has_arg = required_argument},
3538         { .val = LFS_LAYOUT_COPY,
3539                         .name = "copy",         .has_arg = required_argument},
3540         { .val = 'c',   .name = "stripe-count", .has_arg = required_argument},
3541         { .val = 'c',   .name = "stripe_count", .has_arg = required_argument},
3542         { .val = 'c',   .name = "mdt-count",    .has_arg = required_argument},
3543         { .val = 'C',   .name = "overstripe-count",
3544                                                 .has_arg = required_argument},
3545         { .val = 'd',   .name = "delete",       .has_arg = no_argument},
3546         { .val = 'd',   .name = "destroy",      .has_arg = no_argument},
3547         /* --non-direct is only valid in migrate mode */
3548         { .val = 'D',   .name = "non-direct",   .has_arg = no_argument },
3549         { .val = 'E',   .name = "comp-end",     .has_arg = required_argument},
3550         { .val = 'E',   .name = "component-end",
3551                                                 .has_arg = required_argument},
3552         { .val = 'f',   .name = "file",         .has_arg = required_argument },
3553 /* find { .val = 'F',   .name = "fid",          .has_arg = no_argument }, */
3554 /* find { .val = 'g',   .name = "gid",          .has_arg = no_argument }, */
3555 /* find { .val = 'G',   .name = "group",        .has_arg = required_argument }*/
3556         { .val = 'h',   .name = "help",         .has_arg = no_argument },
3557         { .val = 'H',   .name = "mdt-hash",     .has_arg = required_argument},
3558         { .val = 'i',   .name = "stripe-index", .has_arg = required_argument},
3559         { .val = 'i',   .name = "stripe_index", .has_arg = required_argument},
3560         { .val = 'I',   .name = "comp-id",      .has_arg = required_argument},
3561         { .val = 'I',   .name = "component-id", .has_arg = required_argument},
3562 /* find { .val = 'l',   .name = "lazy",         .has_arg = no_argument }, */
3563         { .val = 'L',   .name = "layout",       .has_arg = required_argument },
3564         { .val = 'm',   .name = "mdt",          .has_arg = required_argument},
3565         { .val = 'm',   .name = "mdt-index",    .has_arg = required_argument},
3566         { .val = 'm',   .name = "mdt_index",    .has_arg = required_argument},
3567         /* --non-block is only valid in migrate mode */
3568         { .val = 'n',   .name = "non-block",    .has_arg = no_argument },
3569         { .val = 'N',   .name = "mirror-count", .has_arg = optional_argument},
3570         { .val = 'o',   .name = "ost",          .has_arg = required_argument },
3571 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
3572         { .val = 'o',   .name = "ost-list",     .has_arg = required_argument },
3573         { .val = 'o',   .name = "ost_list",     .has_arg = required_argument },
3574 #endif
3575         { .val = 'p',   .name = "pool",         .has_arg = required_argument },
3576 /* find { .val = 'P',   .name = "print",        .has_arg = no_argument }, */
3577 /* getstripe { .val = 'q', .name = "quiet",     .has_arg = no_argument }, */
3578 /* getstripe { .val = 'r', .name = "recursive", .has_arg = no_argument }, */
3579 /* getstripe { .val = 'R', .name = "raw",       .has_arg = no_argument }, */
3580         { .val = 'S',   .name = "stripe-size",  .has_arg = required_argument },
3581         { .val = 'S',   .name = "stripe_size",  .has_arg = required_argument },
3582 /* find { .val = 't',   .name = "type",         .has_arg = required_argument }*/
3583 /* dirstripe { .val = 'T', .name = "mdt-count", .has_arg = required_argument }*/
3584 /* find { .val = 'u',   .name = "uid",          .has_arg = required_argument }*/
3585 /* find { .val = 'U',   .name = "user",         .has_arg = required_argument }*/
3586         /* --verbose is only valid in migrate mode */
3587         { .val = 'v',   .name = "verbose",      .has_arg = no_argument},
3588         { .val = 'x',   .name = "xattr",        .has_arg = required_argument },
3589 /* dirstripe { .val = 'X',.name = "max-inherit",.has_arg = required_argument }*/
3590         { .val = 'y',   .name = "yaml",         .has_arg = required_argument },
3591         { .val = 'z',   .name = "ext-size",     .has_arg = required_argument},
3592         { .val = 'z',   .name = "extension-size", .has_arg = required_argument},
3593         { .name = NULL } };
3594
3595         setstripe_args_init(&lsa);
3596
3597         migrate_mode = (opc == SO_MIGRATE);
3598         mirror_mode = (opc == SO_MIRROR_CREATE || opc == SO_MIRROR_EXTEND);