kernel/uaudio: Fix weird indent.
[dragonfly.git] / sbin / newfs_hammer2 / newfs_hammer2.c
blobc65e72154f063264047b0ccb4f0df9c589386103
1 /*
2 * Copyright (c) 2011-2014 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
35 #include <sys/types.h>
36 #include <sys/diskslice.h>
37 #include <sys/diskmbr.h>
38 #include <sys/stat.h>
39 #include <sys/time.h>
40 #include <sys/sysctl.h>
41 #include <vfs/hammer2/hammer2_disk.h>
43 #include <stdio.h>
44 #include <stdlib.h>
45 #include <stdarg.h>
46 #include <stddef.h>
47 #include <unistd.h>
48 #include <string.h>
49 #include <errno.h>
50 #include <fcntl.h>
51 #include <assert.h>
52 #include <err.h>
53 #include <uuid.h>
55 #define MAXLABELS HAMMER2_SET_COUNT
57 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
58 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
59 uint32_t iscsi_crc32(const void *buf, size_t size);
60 uint32_t iscsi_crc32_ext(const void *buf, size_t size, uint32_t ocrc);
62 static hammer2_off_t check_volume(const char *path, int *fdp);
63 static int64_t getsize(const char *str, int64_t minval, int64_t maxval, int pw);
64 static const char *sizetostr(hammer2_off_t size);
65 static uint64_t nowtime(void);
66 static int blkrefary_cmp(const void *b1, const void *b2);
67 static void usage(void);
69 static void format_hammer2(int fd, hammer2_off_t total_space,
70 hammer2_off_t free_space);
71 static void alloc_direct(hammer2_off_t *basep, hammer2_blockref_t *bref,
72 size_t bytes);
73 static hammer2_key_t dirhash(const unsigned char *name, size_t len);
75 static int Hammer2Version = -1;
76 static int ForceOpt = 0;
77 static uuid_t Hammer2_FSType; /* static filesystem type id for HAMMER2 */
78 static uuid_t Hammer2_VolFSID; /* unique filesystem id in volu header */
79 static uuid_t Hammer2_SupCLID; /* PFS cluster id in super-root inode */
80 static uuid_t Hammer2_SupFSID; /* PFS unique id in super-root inode */
81 static uuid_t Hammer2_PfsCLID[MAXLABELS];
82 static uuid_t Hammer2_PfsFSID[MAXLABELS];
83 static const char *Label[MAXLABELS];
84 static hammer2_off_t BootAreaSize;
85 static hammer2_off_t AuxAreaSize;
86 static int NLabels;
88 #define GIG ((hammer2_off_t)1024*1024*1024)
90 int
91 main(int ac, char **av)
93 uint32_t status;
94 hammer2_off_t total_space;
95 hammer2_off_t free_space;
96 hammer2_off_t reserved_space;
97 int ch;
98 int fd = -1;
99 int i;
100 int nolabels = 0;
101 char *vol_fsid;
102 char *sup_clid_name;
103 char *sup_fsid_name;
104 char *pfs_clid_name;
105 char *pfs_fsid_name;
107 Label[NLabels++] = "LOCAL";
110 * Sanity check basic filesystem structures. No cookies for us
111 * if it gets broken!
113 assert(sizeof(hammer2_volume_data_t) == HAMMER2_VOLUME_BYTES);
114 assert(sizeof(hammer2_inode_data_t) == HAMMER2_INODE_BYTES);
115 assert(sizeof(hammer2_blockref_t) == HAMMER2_BLOCKREF_BYTES);
118 * Generate a filesystem id and lookup the filesystem type
120 srandomdev();
121 uuidgen(&Hammer2_VolFSID, 1);
122 uuidgen(&Hammer2_SupCLID, 1);
123 uuidgen(&Hammer2_SupFSID, 1);
124 uuid_from_string(HAMMER2_UUID_STRING, &Hammer2_FSType, &status);
125 /*uuid_name_lookup(&Hammer2_FSType, "DragonFly HAMMER2", &status);*/
126 if (status != uuid_s_ok) {
127 errx(1, "uuids file does not have the DragonFly "
128 "HAMMER2 filesystem type");
132 * Parse arguments
134 while ((ch = getopt(ac, av, "fL:b:m:r:V:")) != -1) {
135 switch(ch) {
136 case 'f':
137 ForceOpt = 1;
138 break;
139 case 'L':
140 if (strcasecmp(optarg, "none") == 0) {
141 nolabels = 1;
142 break;
144 if (NLabels >= MAXLABELS) {
145 errx(1,
146 "Limit of %d local labels",
147 MAXLABELS - 1);
149 Label[NLabels++] = optarg;
150 if (strlen(Label[NLabels-1]) > HAMMER2_INODE_MAXNAME) {
151 errx(1, "Root directory label too long "
152 "(64 chars max)\n");
154 break;
155 case 'b':
156 BootAreaSize = getsize(optarg,
157 HAMMER2_NEWFS_ALIGN,
158 HAMMER2_BOOT_MAX_BYTES, 2);
159 break;
160 case 'r':
161 AuxAreaSize = getsize(optarg,
162 HAMMER2_NEWFS_ALIGN,
163 HAMMER2_REDO_MAX_BYTES, 2);
164 break;
165 case 'V':
166 Hammer2Version = strtol(optarg, NULL, 0);
167 if (Hammer2Version < HAMMER2_VOL_VERSION_MIN ||
168 Hammer2Version >= HAMMER2_VOL_VERSION_WIP) {
169 errx(1,
170 "I don't understand how to format "
171 "HAMMER2 version %d\n",
172 Hammer2Version);
174 break;
175 default:
176 usage();
177 break;
182 * Adjust Label[] and NLabels
184 if (nolabels) {
185 NLabels = 1;
186 } else if (NLabels == 1) {
187 Label[NLabels++] = "BOOT";
188 Label[NLabels++] = "ROOT";
192 * Check Hammer2 version
194 if (Hammer2Version < 0) {
195 size_t olen = sizeof(Hammer2Version);
196 Hammer2Version = HAMMER2_VOL_VERSION_DEFAULT;
197 if (sysctlbyname("vfs.hammer2.supported_version",
198 &Hammer2Version, &olen, NULL, 0) == 0) {
199 if (Hammer2Version >= HAMMER2_VOL_VERSION_WIP) {
200 Hammer2Version = HAMMER2_VOL_VERSION_WIP - 1;
201 fprintf(stderr,
202 "newfs_hammer: WARNING: HAMMER2 VFS "
203 "supports higher version than I "
204 "understand,\n"
205 "using version %d\n",
206 Hammer2Version);
208 } else {
209 fprintf(stderr,
210 "newfs_hammer: WARNING: HAMMER2 VFS not "
211 "loaded, cannot get version info.\n"
212 "Using version %d\n",
213 HAMMER2_VOL_VERSION_DEFAULT);
218 * Collect volume information.
220 ac -= optind;
221 av += optind;
223 if (ac != 1) {
224 fprintf(stderr, "Exactly one disk device must be specified\n");
225 exit(1);
227 total_space = check_volume(av[0], &fd);
230 * ~typically 8MB alignment to avoid edge cases for reserved blocks
231 * and so raid stripes (if any) operate efficiently.
233 total_space &= ~HAMMER2_VOLUME_ALIGNMASK64;
236 * Calculate defaults for the boot area size and round to the
237 * volume alignment boundary.
239 if (BootAreaSize == 0) {
240 BootAreaSize = HAMMER2_BOOT_NOM_BYTES;
241 while (BootAreaSize > total_space / 20)
242 BootAreaSize >>= 1;
243 if (BootAreaSize < HAMMER2_BOOT_MIN_BYTES)
244 BootAreaSize = HAMMER2_BOOT_MIN_BYTES;
245 } else if (BootAreaSize < HAMMER2_BOOT_MIN_BYTES) {
246 BootAreaSize = HAMMER2_BOOT_MIN_BYTES;
248 BootAreaSize = (BootAreaSize + HAMMER2_VOLUME_ALIGNMASK64) &
249 ~HAMMER2_VOLUME_ALIGNMASK64;
252 * Calculate defaults for the redo area size and round to the
253 * volume alignment boundary.
255 if (AuxAreaSize == 0) {
256 AuxAreaSize = HAMMER2_REDO_NOM_BYTES;
257 while (AuxAreaSize > total_space / 20)
258 AuxAreaSize >>= 1;
259 if (AuxAreaSize < HAMMER2_REDO_MIN_BYTES)
260 AuxAreaSize = HAMMER2_REDO_MIN_BYTES;
261 } else if (AuxAreaSize < HAMMER2_REDO_MIN_BYTES) {
262 AuxAreaSize = HAMMER2_REDO_MIN_BYTES;
264 AuxAreaSize = (AuxAreaSize + HAMMER2_VOLUME_ALIGNMASK64) &
265 ~HAMMER2_VOLUME_ALIGNMASK64;
268 * We'll need to stuff this in the volume header soon.
270 uuid_to_string(&Hammer2_VolFSID, &vol_fsid, &status);
271 uuid_to_string(&Hammer2_SupCLID, &sup_clid_name, &status);
272 uuid_to_string(&Hammer2_SupFSID, &sup_fsid_name, &status);
275 * Calculate the amount of reserved space. HAMMER2_ZONE_SEG (4MB)
276 * is reserved at the beginning of every 2GB of storage, rounded up.
277 * Thus a 200MB filesystem will still have a 4MB reserve area.
279 * We also include the boot and redo areas in the reserve. The
280 * reserve is used to help 'df' calculate the amount of available
281 * space.
283 reserved_space = ((total_space + HAMMER2_ZONE_MASK64) /
284 HAMMER2_ZONE_BYTES64) * HAMMER2_ZONE_SEG64;
286 free_space = total_space - reserved_space -
287 BootAreaSize - AuxAreaSize;
289 format_hammer2(fd, total_space, free_space);
290 fsync(fd);
291 close(fd);
293 printf("---------------------------------------------\n");
294 printf("version: %d\n", Hammer2Version);
295 printf("total-size: %s (%jd bytes)\n",
296 sizetostr(total_space),
297 (intmax_t)total_space);
298 printf("boot-area-size: %s\n", sizetostr(BootAreaSize));
299 printf("aux-area-size: %s\n", sizetostr(AuxAreaSize));
300 printf("topo-reserved: %s\n", sizetostr(reserved_space));
301 printf("free-space: %s\n", sizetostr(free_space));
302 printf("vol-fsid: %s\n", vol_fsid);
303 printf("sup-clid: %s\n", sup_clid_name);
304 printf("sup-fsid: %s\n", sup_fsid_name);
305 for (i = 0; i < NLabels; ++i) {
306 printf("PFS \"%s\"\n", Label[i]);
307 uuid_to_string(&Hammer2_PfsCLID[i], &pfs_clid_name, &status);
308 uuid_to_string(&Hammer2_PfsFSID[i], &pfs_fsid_name, &status);
309 printf(" clid %s\n", pfs_clid_name);
310 printf(" fsid %s\n", pfs_fsid_name);
312 printf("\n");
314 return(0);
317 static
318 void
319 usage(void)
321 fprintf(stderr,
322 "usage: newfs_hammer -L label [-f] [-b bootsize] "
323 "[-r redosize] [-V version] special ...\n"
325 exit(1);
329 * Convert the size in bytes to a human readable string.
331 static
332 const char *
333 sizetostr(hammer2_off_t size)
335 static char buf[32];
337 if (size < 1024 / 2) {
338 snprintf(buf, sizeof(buf), "%6.2f", (double)size);
339 } else if (size < 1024 * 1024 / 2) {
340 snprintf(buf, sizeof(buf), "%6.2fKB",
341 (double)size / 1024);
342 } else if (size < 1024 * 1024 * 1024LL / 2) {
343 snprintf(buf, sizeof(buf), "%6.2fMB",
344 (double)size / (1024 * 1024));
345 } else if (size < 1024 * 1024 * 1024LL * 1024LL / 2) {
346 snprintf(buf, sizeof(buf), "%6.2fGB",
347 (double)size / (1024 * 1024 * 1024LL));
348 } else {
349 snprintf(buf, sizeof(buf), "%6.2fTB",
350 (double)size / (1024 * 1024 * 1024LL * 1024LL));
352 return(buf);
356 * Convert a string to a 64 bit signed integer with various requirements.
358 static int64_t
359 getsize(const char *str, int64_t minval, int64_t maxval, int powerof2)
361 int64_t val;
362 char *ptr;
364 val = strtoll(str, &ptr, 0);
365 switch(*ptr) {
366 case 't':
367 case 'T':
368 val *= 1024;
369 /* fall through */
370 case 'g':
371 case 'G':
372 val *= 1024;
373 /* fall through */
374 case 'm':
375 case 'M':
376 val *= 1024;
377 /* fall through */
378 case 'k':
379 case 'K':
380 val *= 1024;
381 break;
382 default:
383 errx(1, "Unknown suffix in number '%s'\n", str);
384 /* not reached */
386 if (ptr[1]) {
387 errx(1, "Unknown suffix in number '%s'\n", str);
388 /* not reached */
390 if (val < minval) {
391 errx(1, "Value too small: %s, min is %s\n",
392 str, sizetostr(minval));
393 /* not reached */
395 if (val > maxval) {
396 errx(1, "Value too large: %s, max is %s\n",
397 str, sizetostr(maxval));
398 /* not reached */
400 if ((powerof2 & 1) && (val ^ (val - 1)) != ((val << 1) - 1)) {
401 errx(1, "Value not power of 2: %s\n", str);
402 /* not reached */
404 if ((powerof2 & 2) && (val & HAMMER2_NEWFS_ALIGNMASK)) {
405 errx(1, "Value not an integral multiple of %dK: %s",
406 HAMMER2_NEWFS_ALIGN / 1024, str);
407 /* not reached */
409 return(val);
412 static uint64_t
413 nowtime(void)
415 struct timeval tv;
416 uint64_t xtime;
418 gettimeofday(&tv, NULL);
419 xtime = tv.tv_sec * 1000000LL + tv.tv_usec;
420 return(xtime);
424 * Figure out how big the volume is.
426 static
427 hammer2_off_t
428 check_volume(const char *path, int *fdp)
430 struct partinfo pinfo;
431 struct stat st;
432 hammer2_off_t size;
435 * Get basic information about the volume
437 *fdp = open(path, O_RDWR);
438 if (*fdp < 0)
439 err(1, "Unable to open %s R+W", path);
440 if (ioctl(*fdp, DIOCGPART, &pinfo) < 0) {
442 * Allow the formatting of regular files as HAMMER2 volumes
444 if (fstat(*fdp, &st) < 0)
445 err(1, "Unable to stat %s", path);
446 size = st.st_size;
447 } else {
449 * When formatting a block device as a HAMMER2 volume the
450 * sector size must be compatible. HAMMER2 uses 64K
451 * filesystem buffers but logical buffers for direct I/O
452 * can be as small as HAMMER2_LOGSIZE (16KB).
454 if (pinfo.reserved_blocks) {
455 errx(1, "HAMMER cannot be placed in a partition "
456 "which overlaps the disklabel or MBR");
458 if (pinfo.media_blksize > HAMMER2_PBUFSIZE ||
459 HAMMER2_PBUFSIZE % pinfo.media_blksize) {
460 errx(1, "A media sector size of %d is not supported",
461 pinfo.media_blksize);
463 size = pinfo.media_size;
465 printf("Volume %-15s size %s\n", path, sizetostr(size));
466 return (size);
470 * Create the volume header, the super-root directory inode, and
471 * the writable snapshot subdirectory (named via the label) which
472 * is to be the initial mount point, or at least the first mount point.
474 * [----reserved_area----][boot_area][aux_area]
475 * [[vol_hdr]... ] [sroot][root]
477 * The sroot and root inodes eat 512 bytes each. newfs labels can only be
478 * 64 bytes so the root (snapshot) inode does not need to extend past 512
479 * bytes. We use the correct hash slot correct but note that because
480 * directory hashes are chained 16x, any slot in the inode will work.
482 * Also format the allocation map.
484 * NOTE: The passed total_space is 8MB-aligned to avoid edge cases.
486 static
487 void
488 format_hammer2(int fd, hammer2_off_t total_space, hammer2_off_t free_space)
490 char *buf = malloc(HAMMER2_PBUFSIZE);
491 hammer2_volume_data_t *vol;
492 hammer2_inode_data_t *rawip;
493 hammer2_blockref_t sroot_blockref;
494 hammer2_blockref_t root_blockref[MAXLABELS];
495 uint64_t now;
496 hammer2_off_t volu_base = 0;
497 hammer2_off_t boot_base = HAMMER2_ZONE_SEG;
498 hammer2_off_t aux_base = boot_base + BootAreaSize;
499 hammer2_off_t alloc_base = aux_base + AuxAreaSize;
500 hammer2_off_t tmp_base;
501 size_t n;
502 int i;
505 * Clear the entire reserve for the first 2G segment and
506 * make sure we can write to the last block.
508 bzero(buf, HAMMER2_PBUFSIZE);
509 tmp_base = volu_base;
510 for (i = 0; i < HAMMER2_ZONE_BLOCKS_SEG; ++i) {
511 n = pwrite(fd, buf, HAMMER2_PBUFSIZE, tmp_base);
512 if (n != HAMMER2_PBUFSIZE) {
513 perror("write");
514 exit(1);
516 tmp_base += HAMMER2_PBUFSIZE;
519 n = pwrite(fd, buf, HAMMER2_PBUFSIZE,
520 volu_base + total_space - HAMMER2_PBUFSIZE);
521 if (n != HAMMER2_PBUFSIZE) {
522 perror("write (at-end-of-volume)");
523 exit(1);
527 * Make sure alloc_base won't cross the reserved area at the
528 * beginning of each 2GB zone.
530 * Reserve space for the super-root inode and the root inode.
531 * Make sure they are in the same 64K block to simplify our code.
533 assert((alloc_base & HAMMER2_PBUFMASK) == 0);
534 assert(alloc_base < HAMMER2_ZONE_BYTES64 - HAMMER2_ZONE_SEG);
535 now = nowtime();
536 bzero(buf, HAMMER2_PBUFSIZE);
538 alloc_base &= ~HAMMER2_PBUFMASK64;
539 alloc_direct(&alloc_base, &sroot_blockref, HAMMER2_INODE_BYTES);
541 for (i = 0; i < NLabels; ++i) {
542 uuidgen(&Hammer2_PfsCLID[i], 1);
543 uuidgen(&Hammer2_PfsFSID[i], 1);
545 alloc_direct(&alloc_base, &root_blockref[i],
546 HAMMER2_INODE_BYTES);
547 assert(((sroot_blockref.data_off ^ root_blockref[i].data_off) &
548 HAMMER2_OFF_MASK_HI) == 0);
551 * Format the root directory inode, which is left empty.
553 rawip = (void *)(buf + (HAMMER2_OFF_MASK_LO &
554 root_blockref[i].data_off));
555 rawip->meta.version = HAMMER2_INODE_VERSION_ONE;
556 rawip->meta.ctime = now;
557 rawip->meta.mtime = now;
558 /* rawip->atime = now; NOT IMPL MUST BE ZERO */
559 rawip->meta.btime = now;
560 rawip->meta.type = HAMMER2_OBJTYPE_DIRECTORY;
561 rawip->meta.mode = 0755;
562 rawip->meta.inum = 1; /* root inode, inumber 1 */
563 rawip->meta.nlinks = 1; /* directory link count compat */
565 rawip->meta.name_len = strlen(Label[i]);
566 bcopy(Label[i], rawip->filename, rawip->meta.name_len);
567 rawip->meta.name_key =
568 dirhash(rawip->filename, rawip->meta.name_len);
571 * Compression mode and supported copyids.
573 * Do not allow compression when creating any "BOOT" label
574 * (pfs-create also does the same if the pfs is named "BOOT")
576 if (strcasecmp(Label[i], "BOOT") == 0) {
577 rawip->meta.comp_algo = HAMMER2_ENC_ALGO(
578 HAMMER2_COMP_AUTOZERO);
579 rawip->meta.check_algo = HAMMER2_ENC_ALGO(
580 HAMMER2_CHECK_ISCSI32);
581 } else {
582 rawip->meta.comp_algo = HAMMER2_ENC_ALGO(
583 HAMMER2_COMP_NEWFS_DEFAULT);
584 rawip->meta.check_algo = HAMMER2_ENC_ALGO(
585 HAMMER2_CHECK_ISCSI32);
589 * NOTE: We leave nmasters set to 0, which means that we
590 * don't know how many masters there are. The quorum
591 * calculation will effectively be 1 ( 0 / 2 + 1 ).
593 rawip->meta.pfs_clid = Hammer2_PfsCLID[i];
594 rawip->meta.pfs_fsid = Hammer2_PfsFSID[i];
595 rawip->meta.pfs_type = HAMMER2_PFSTYPE_MASTER;
596 rawip->meta.op_flags |= HAMMER2_OPFLAG_PFSROOT;
598 /* first allocatable inode number */
599 rawip->meta.pfs_inum = 16;
601 /* rawip->u.blockset is left empty */
604 * The root blockref will be stored in the super-root inode as
605 * the only directory entry. The copyid here is the actual
606 * copyid of the storage ref.
608 * The key field for a directory entry's blockref is
609 * essentially the name key for the entry.
611 root_blockref[i].key = rawip->meta.name_key;
612 root_blockref[i].copyid = HAMMER2_COPYID_LOCAL;
613 root_blockref[i].keybits = 0;
614 root_blockref[i].check.iscsi32.value =
615 hammer2_icrc32(rawip, sizeof(*rawip));
616 root_blockref[i].type = HAMMER2_BREF_TYPE_INODE;
617 root_blockref[i].methods =
618 HAMMER2_ENC_CHECK(HAMMER2_CHECK_ISCSI32) |
619 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
620 root_blockref[i].mirror_tid = 16;
621 root_blockref[i].flags = HAMMER2_BREF_FLAG_PFSROOT;
625 * Format the super-root directory inode, giving it one directory
626 * entry (root_blockref) and fixup the icrc method.
628 * The superroot contains one directory entry pointing at the root
629 * inode (named via the label). Inodes contain one blockset which
630 * is fully associative so we can put the entry anywhere without
631 * having to worry about the hash. Use index 0.
633 rawip = (void *)(buf + (HAMMER2_OFF_MASK_LO & sroot_blockref.data_off));
634 rawip->meta.version = HAMMER2_INODE_VERSION_ONE;
635 rawip->meta.ctime = now;
636 rawip->meta.mtime = now;
637 /* rawip->meta.atime = now; NOT IMPL MUST BE ZERO */
638 rawip->meta.btime = now;
639 rawip->meta.type = HAMMER2_OBJTYPE_DIRECTORY;
640 rawip->meta.mode = 0700; /* super-root - root only */
641 rawip->meta.inum = 0; /* super root inode, inumber 0 */
642 rawip->meta.nlinks = 2; /* directory link count compat */
644 rawip->meta.name_len = 0; /* super-root is unnamed */
645 rawip->meta.name_key = 0;
647 rawip->meta.comp_algo = HAMMER2_ENC_ALGO(HAMMER2_COMP_AUTOZERO);
648 rawip->meta.check_algo = HAMMER2_ENC_ALGO(HAMMER2_CHECK_ISCSI32);
651 * The super-root is flagged as a PFS and typically given its own
652 * random FSID, making it possible to mirror an entire HAMMER2 disk
653 * snapshots and all if desired. PFS ids are used to match up
654 * mirror sources and targets and cluster copy sources and targets.
656 * (XXX whole-disk logical mirroring is not really supported in
657 * the first attempt because each PFS is in its own modify/mirror
658 * transaction id domain, so normal mechanics cannot cross a PFS
659 * boundary).
661 rawip->meta.pfs_clid = Hammer2_SupCLID;
662 rawip->meta.pfs_fsid = Hammer2_SupFSID;
663 rawip->meta.pfs_type = HAMMER2_PFSTYPE_SUPROOT;
664 snprintf(rawip->filename, sizeof(rawip->filename), "SUPROOT");
665 rawip->meta.name_key = 0;
666 rawip->meta.name_len = strlen(rawip->filename);
668 /* The super-root has an inode number of 0 */
669 rawip->meta.pfs_inum = 0;
672 * Currently newfs_hammer2 just throws the PFS inodes into the
673 * top-level block table at the volume root and doesn't try to
674 * create an indirect block, so we are limited to ~4 at filesystem
675 * creation time. More can be added after mounting.
677 qsort(root_blockref, NLabels, sizeof(root_blockref[0]), blkrefary_cmp);
678 for (i = 0; i < NLabels; ++i)
679 rawip->u.blockset.blockref[i] = root_blockref[i];
682 * The sroot blockref will be stored in the volume header.
684 sroot_blockref.copyid = HAMMER2_COPYID_LOCAL;
685 sroot_blockref.keybits = 0;
686 sroot_blockref.check.iscsi32.value =
687 hammer2_icrc32(rawip, sizeof(*rawip));
688 sroot_blockref.type = HAMMER2_BREF_TYPE_INODE;
689 sroot_blockref.methods = HAMMER2_ENC_CHECK(HAMMER2_CHECK_ISCSI32) |
690 HAMMER2_ENC_COMP(HAMMER2_COMP_AUTOZERO);
691 sroot_blockref.mirror_tid = 16;
692 rawip = NULL;
695 * Write out the 64K HAMMER2 block containing the root and sroot.
697 n = pwrite(fd, buf, HAMMER2_PBUFSIZE,
698 sroot_blockref.data_off & HAMMER2_OFF_MASK_HI);
699 if (n != HAMMER2_PBUFSIZE) {
700 perror("write");
701 exit(1);
705 * Format the volume header.
707 * The volume header points to sroot_blockref. Also be absolutely
708 * sure that allocator_beg is set.
710 bzero(buf, HAMMER2_PBUFSIZE);
711 vol = (void *)buf;
713 vol->magic = HAMMER2_VOLUME_ID_HBO;
714 vol->boot_beg = boot_base;
715 vol->boot_end = boot_base + BootAreaSize;
716 vol->aux_beg = aux_base;
717 vol->aux_end = aux_base + AuxAreaSize;
718 vol->volu_size = total_space;
719 vol->version = Hammer2Version;
720 vol->flags = 0;
722 vol->fsid = Hammer2_VolFSID;
723 vol->fstype = Hammer2_FSType;
725 vol->peer_type = DMSG_PEER_HAMMER2; /* LNK_CONN identification */
727 vol->allocator_size = free_space;
728 vol->allocator_free = free_space;
729 vol->allocator_beg = alloc_base;
731 vol->sroot_blockset.blockref[0] = sroot_blockref;
732 vol->mirror_tid = 16; /* all blockref mirror TIDs set to 16 */
733 vol->freemap_tid = 16; /* all blockref mirror TIDs set to 16 */
734 vol->icrc_sects[HAMMER2_VOL_ICRC_SECT1] =
735 hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRC1_OFF,
736 HAMMER2_VOLUME_ICRC1_SIZE);
739 * Set ICRC_SECT0 after all remaining elements of sect0 have been
740 * populated in the volume header. Note hat ICRC_SECT* (except for
741 * SECT0) are part of sect0.
743 vol->icrc_sects[HAMMER2_VOL_ICRC_SECT0] =
744 hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRC0_OFF,
745 HAMMER2_VOLUME_ICRC0_SIZE);
746 vol->icrc_volheader =
747 hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRCVH_OFF,
748 HAMMER2_VOLUME_ICRCVH_SIZE);
751 * Write the volume header and all alternates.
753 for (i = 0; i < HAMMER2_NUM_VOLHDRS; ++i) {
754 if (i * HAMMER2_ZONE_BYTES64 >= total_space)
755 break;
756 n = pwrite(fd, buf, HAMMER2_PBUFSIZE,
757 volu_base + i * HAMMER2_ZONE_BYTES64);
758 if (n != HAMMER2_PBUFSIZE) {
759 perror("write");
760 exit(1);
765 * Cleanup
767 free(buf);
770 static void
771 alloc_direct(hammer2_off_t *basep, hammer2_blockref_t *bref, size_t bytes)
773 int radix;
775 radix = 0;
776 assert(bytes);
777 while ((bytes & 1) == 0) {
778 bytes >>= 1;
779 ++radix;
781 assert(bytes == 1);
782 if (radix < HAMMER2_RADIX_MIN)
783 radix = HAMMER2_RADIX_MIN;
785 bzero(bref, sizeof(*bref));
786 bref->data_off = *basep | radix;
787 bref->vradix = radix;
789 *basep += 1U << radix;
793 * Borrow HAMMER1's directory hash algorithm #1 with a few modifications.
794 * The filename is split into fields which are hashed separately and then
795 * added together.
797 * Differences include: bit 63 must be set to 1 for HAMMER2 (HAMMER1 sets
798 * it to 0), this is because bit63=0 is used for hidden hardlinked inodes.
799 * (This means we do not need to do a 0-check/or-with-0x100000000 either).
801 * Also, the iscsi crc code is used instead of the old crc32 code.
803 static hammer2_key_t
804 dirhash(const unsigned char *name, size_t len)
806 const unsigned char *aname = name;
807 uint32_t crcx;
808 uint64_t key;
809 size_t i;
810 size_t j;
813 * Filesystem version 6 or better will create directories
814 * using the ALG1 dirhash. This hash breaks the filename
815 * up into domains separated by special characters and
816 * hashes each domain independently.
818 * We also do a simple sub-sort using the first character
819 * of the filename in the top 5-bits.
821 key = 0;
824 * m32
826 crcx = 0;
827 for (i = j = 0; i < len; ++i) {
828 if (aname[i] == '.' ||
829 aname[i] == '-' ||
830 aname[i] == '_' ||
831 aname[i] == '~') {
832 if (i != j)
833 crcx += hammer2_icrc32(aname + j, i - j);
834 j = i + 1;
837 if (i != j)
838 crcx += hammer2_icrc32(aname + j, i - j);
841 * The directory hash utilizes the top 32 bits of the 64-bit key.
842 * Bit 63 must be set to 1.
844 crcx |= 0x80000000U;
845 key |= (uint64_t)crcx << 32;
848 * l16 - crc of entire filename
850 * This crc reduces degenerate hash collision conditions
852 crcx = hammer2_icrc32(aname, len);
853 crcx = crcx ^ (crcx << 16);
854 key |= crcx & 0xFFFF0000U;
857 * Set bit 15. This allows readdir to strip bit 63 so a positive
858 * 64-bit cookie/offset can always be returned, and still guarantee
859 * that the values 0x0000-0x7FFF are available for artificial entries.
860 * ('.' and '..').
862 key |= 0x8000U;
864 return (key);
867 static int
868 blkrefary_cmp(const void *b1, const void *b2)
870 const hammer2_blockref_t *bref1 = b1;
871 const hammer2_blockref_t *bref2 = b2;
872 if (bref1->key < bref2->key)
873 return(-1);
874 if (bref1->key > bref2->key)
875 return(1);
876 return 0;