xref: /qemu/pc-bios/s390-ccw/bootmap.c (revision a81df1b6)
1 /*
2  * QEMU S390 bootmap interpreter
3  *
4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or (at
7  * your option) any later version. See the COPYING file in the top-level
8  * directory.
9  */
10 
11 #include "libc.h"
12 #include "s390-ccw.h"
13 #include "bootmap.h"
14 #include "virtio.h"
15 #include "bswap.h"
16 
17 #ifdef DEBUG
18 /* #define DEBUG_FALLBACK */
19 #endif
20 
21 #ifdef DEBUG_FALLBACK
22 #define dputs(txt) \
23     do { sclp_print("zipl: " txt); } while (0)
24 #else
25 #define dputs(fmt, ...) \
26     do { } while (0)
27 #endif
28 
29 /* Scratch space */
30 static uint8_t sec[MAX_SECTOR_SIZE*4] __attribute__((__aligned__(PAGE_SIZE)));
31 
32 const uint8_t el_torito_magic[] = "EL TORITO SPECIFICATION"
33                                   "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
34 
35 /*
36  * Match two CCWs located after PSW and eight filler bytes.
37  * From libmagic and arch/s390/kernel/head.S.
38  */
39 const uint8_t linux_s390_magic[] = "\x02\x00\x00\x18\x60\x00\x00\x50\x02\x00"
40                                    "\x00\x68\x60\x00\x00\x50\x40\x40\x40\x40"
41                                    "\x40\x40\x40\x40";
42 
43 static inline bool is_iso_vd_valid(IsoVolDesc *vd)
44 {
45     const uint8_t vol_desc_magic[] = "CD001";
46 
47     return !memcmp(&vd->ident[0], vol_desc_magic, 5) &&
48            vd->version == 0x1 &&
49            vd->type <= VOL_DESC_TYPE_PARTITION;
50 }
51 
52 /***********************************************************************
53  * IPL an ECKD DASD (CDL or LDL/CMS format)
54  */
55 
56 static unsigned char _bprs[8*1024]; /* guessed "max" ECKD sector size */
57 static const int max_bprs_entries = sizeof(_bprs) / sizeof(ExtEckdBlockPtr);
58 static uint8_t _s2[MAX_SECTOR_SIZE * 3] __attribute__((__aligned__(PAGE_SIZE)));
59 static void *s2_prev_blk = _s2;
60 static void *s2_cur_blk = _s2 + MAX_SECTOR_SIZE;
61 static void *s2_next_blk = _s2 + MAX_SECTOR_SIZE * 2;
62 
63 static inline void verify_boot_info(BootInfo *bip)
64 {
65     IPL_assert(magic_match(bip->magic, ZIPL_MAGIC), "No zIPL sig in BootInfo");
66     IPL_assert(bip->version == BOOT_INFO_VERSION, "Wrong zIPL version");
67     IPL_assert(bip->bp_type == BOOT_INFO_BP_TYPE_IPL, "DASD is not for IPL");
68     IPL_assert(bip->dev_type == BOOT_INFO_DEV_TYPE_ECKD, "DASD is not ECKD");
69     IPL_assert(bip->flags == BOOT_INFO_FLAGS_ARCH, "Not for this arch");
70     IPL_assert(block_size_ok(bip->bp.ipl.bm_ptr.eckd.bptr.size),
71                "Bad block size in zIPL section of the 1st record.");
72 }
73 
74 static block_number_t eckd_block_num(EckdCHS *chs)
75 {
76     const uint64_t sectors = virtio_get_sectors();
77     const uint64_t heads = virtio_get_heads();
78     const uint64_t cylinder = chs->cylinder
79                             + ((chs->head & 0xfff0) << 12);
80     const uint64_t head = chs->head & 0x000f;
81     const block_number_t block = sectors * heads * cylinder
82                                + sectors * head
83                                + chs->sector
84                                - 1; /* block nr starts with zero */
85     return block;
86 }
87 
88 static bool eckd_valid_address(BootMapPointer *p)
89 {
90     const uint64_t head = p->eckd.chs.head & 0x000f;
91 
92     if (head >= virtio_get_heads()
93         ||  p->eckd.chs.sector > virtio_get_sectors()
94         ||  p->eckd.chs.sector <= 0) {
95         return false;
96     }
97 
98     if (!virtio_guessed_disk_nature() &&
99         eckd_block_num(&p->eckd.chs) >= virtio_get_blocks()) {
100         return false;
101     }
102 
103     return true;
104 }
105 
106 static block_number_t load_eckd_segments(block_number_t blk, uint64_t *address)
107 {
108     block_number_t block_nr;
109     int j, rc;
110     BootMapPointer *bprs = (void *)_bprs;
111     bool more_data;
112 
113     memset(_bprs, FREE_SPACE_FILLER, sizeof(_bprs));
114     read_block(blk, bprs, "BPRS read failed");
115 
116     do {
117         more_data = false;
118         for (j = 0;; j++) {
119             block_nr = eckd_block_num(&bprs[j].xeckd.bptr.chs);
120             if (is_null_block_number(block_nr)) { /* end of chunk */
121                 break;
122             }
123 
124             /* we need the updated blockno for the next indirect entry
125              * in the chain, but don't want to advance address
126              */
127             if (j == (max_bprs_entries - 1)) {
128                 break;
129             }
130 
131             IPL_assert(block_size_ok(bprs[j].xeckd.bptr.size),
132                        "bad chunk block size");
133             IPL_assert(eckd_valid_address(&bprs[j]), "bad chunk ECKD addr");
134 
135             if ((bprs[j].xeckd.bptr.count == 0) && unused_space(&(bprs[j+1]),
136                 sizeof(EckdBlockPtr))) {
137                 /* This is a "continue" pointer.
138                  * This ptr should be the last one in the current
139                  * script section.
140                  * I.e. the next ptr must point to the unused memory area
141                  */
142                 memset(_bprs, FREE_SPACE_FILLER, sizeof(_bprs));
143                 read_block(block_nr, bprs, "BPRS continuation read failed");
144                 more_data = true;
145                 break;
146             }
147 
148             /* Load (count+1) blocks of code at (block_nr)
149              * to memory (address).
150              */
151             rc = virtio_read_many(block_nr, (void *)(*address),
152                                   bprs[j].xeckd.bptr.count+1);
153             IPL_assert(rc == 0, "code chunk read failed");
154 
155             *address += (bprs[j].xeckd.bptr.count+1) * virtio_get_block_size();
156         }
157     } while (more_data);
158     return block_nr;
159 }
160 
161 static bool find_zipl_boot_menu_banner(int *offset)
162 {
163     int i;
164 
165     /* Menu banner starts with "zIPL" */
166     for (i = 0; i < virtio_get_block_size() - 4; i++) {
167         if (magic_match(s2_cur_blk + i, ZIPL_MAGIC_EBCDIC)) {
168             *offset = i;
169             return true;
170         }
171     }
172 
173     return false;
174 }
175 
176 static int eckd_get_boot_menu_index(block_number_t s1b_block_nr)
177 {
178     block_number_t cur_block_nr;
179     block_number_t prev_block_nr = 0;
180     block_number_t next_block_nr = 0;
181     EckdStage1b *s1b = (void *)sec;
182     int banner_offset;
183     int i;
184 
185     /* Get Stage1b data */
186     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
187     read_block(s1b_block_nr, s1b, "Cannot read stage1b boot loader");
188 
189     memset(_s2, FREE_SPACE_FILLER, sizeof(_s2));
190 
191     /* Get Stage2 data */
192     for (i = 0; i < STAGE2_BLK_CNT_MAX; i++) {
193         cur_block_nr = eckd_block_num(&s1b->seek[i].chs);
194 
195         if (!cur_block_nr) {
196             break;
197         }
198 
199         read_block(cur_block_nr, s2_cur_blk, "Cannot read stage2 boot loader");
200 
201         if (find_zipl_boot_menu_banner(&banner_offset)) {
202             /*
203              * Load the adjacent blocks to account for the
204              * possibility of menu data spanning multiple blocks.
205              */
206             if (prev_block_nr) {
207                 read_block(prev_block_nr, s2_prev_blk,
208                            "Cannot read stage2 boot loader");
209             }
210 
211             if (i + 1 < STAGE2_BLK_CNT_MAX) {
212                 next_block_nr = eckd_block_num(&s1b->seek[i + 1].chs);
213             }
214 
215             if (next_block_nr) {
216                 read_block(next_block_nr, s2_next_blk,
217                            "Cannot read stage2 boot loader");
218             }
219 
220             return menu_get_zipl_boot_index(s2_cur_blk + banner_offset);
221         }
222 
223         prev_block_nr = cur_block_nr;
224     }
225 
226     sclp_print("No zipl boot menu data found. Booting default entry.");
227     return 0;
228 }
229 
230 static void run_eckd_boot_script(block_number_t bmt_block_nr,
231                                  block_number_t s1b_block_nr)
232 {
233     int i;
234     unsigned int loadparm = get_loadparm_index();
235     block_number_t block_nr;
236     uint64_t address;
237     BootMapTable *bmt = (void *)sec;
238     BootMapScript *bms = (void *)sec;
239 
240     if (menu_is_enabled_zipl()) {
241         loadparm = eckd_get_boot_menu_index(s1b_block_nr);
242     }
243 
244     debug_print_int("loadparm", loadparm);
245     IPL_assert(loadparm < MAX_BOOT_ENTRIES, "loadparm value greater than"
246                " maximum number of boot entries allowed");
247 
248     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
249     read_block(bmt_block_nr, sec, "Cannot read Boot Map Table");
250 
251     block_nr = eckd_block_num(&bmt->entry[loadparm].xeckd.bptr.chs);
252     IPL_assert(block_nr != -1, "Cannot find Boot Map Table Entry");
253 
254     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
255     read_block(block_nr, sec, "Cannot read Boot Map Script");
256 
257     for (i = 0; bms->entry[i].type == BOOT_SCRIPT_LOAD ||
258                 bms->entry[i].type == BOOT_SCRIPT_SIGNATURE; i++) {
259 
260         /* We don't support secure boot yet, so we skip signature entries */
261         if (bms->entry[i].type == BOOT_SCRIPT_SIGNATURE) {
262             continue;
263         }
264 
265         address = bms->entry[i].address.load_address;
266         block_nr = eckd_block_num(&bms->entry[i].blkptr.xeckd.bptr.chs);
267 
268         do {
269             block_nr = load_eckd_segments(block_nr, &address);
270         } while (block_nr != -1);
271     }
272 
273     IPL_assert(bms->entry[i].type == BOOT_SCRIPT_EXEC,
274                "Unknown script entry type");
275     jump_to_IPL_code(bms->entry[i].address.load_address); /* no return */
276 }
277 
278 static void ipl_eckd_cdl(void)
279 {
280     XEckdMbr *mbr;
281     EckdCdlIpl2 *ipl2 = (void *)sec;
282     IplVolumeLabel *vlbl = (void *)sec;
283     block_number_t bmt_block_nr, s1b_block_nr;
284 
285     /* we have just read the block #0 and recognized it as "IPL1" */
286     sclp_print("CDL\n");
287 
288     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
289     read_block(1, ipl2, "Cannot read IPL2 record at block 1");
290 
291     mbr = &ipl2->mbr;
292     IPL_assert(magic_match(mbr, ZIPL_MAGIC), "No zIPL section in IPL2 record.");
293     IPL_assert(block_size_ok(mbr->blockptr.xeckd.bptr.size),
294                "Bad block size in zIPL section of IPL2 record.");
295     IPL_assert(mbr->dev_type == DEV_TYPE_ECKD,
296                "Non-ECKD device type in zIPL section of IPL2 record.");
297 
298     /* save pointer to Boot Map Table */
299     bmt_block_nr = eckd_block_num(&mbr->blockptr.xeckd.bptr.chs);
300 
301     /* save pointer to Stage1b Data */
302     s1b_block_nr = eckd_block_num(&ipl2->stage1.seek[0].chs);
303 
304     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
305     read_block(2, vlbl, "Cannot read Volume Label at block 2");
306     IPL_assert(magic_match(vlbl->key, VOL1_MAGIC),
307                "Invalid magic of volume label block");
308     IPL_assert(magic_match(vlbl->f.key, VOL1_MAGIC),
309                "Invalid magic of volser block");
310     print_volser(vlbl->f.volser);
311 
312     run_eckd_boot_script(bmt_block_nr, s1b_block_nr);
313     /* no return */
314 }
315 
316 static void print_eckd_ldl_msg(ECKD_IPL_mode_t mode)
317 {
318     LDL_VTOC *vlbl = (void *)sec; /* already read, 3rd block */
319     char msg[4] = { '?', '.', '\n', '\0' };
320 
321     sclp_print((mode == ECKD_CMS) ? "CMS" : "LDL");
322     sclp_print(" version ");
323     switch (vlbl->LDL_version) {
324     case LDL1_VERSION:
325         msg[0] = '1';
326         break;
327     case LDL2_VERSION:
328         msg[0] = '2';
329         break;
330     default:
331         msg[0] = ebc2asc[vlbl->LDL_version];
332         msg[1] = '?';
333         break;
334     }
335     sclp_print(msg);
336     print_volser(vlbl->volser);
337 }
338 
339 static void ipl_eckd_ldl(ECKD_IPL_mode_t mode)
340 {
341     block_number_t bmt_block_nr, s1b_block_nr;
342     EckdLdlIpl1 *ipl1 = (void *)sec;
343 
344     if (mode != ECKD_LDL_UNLABELED) {
345         print_eckd_ldl_msg(mode);
346     }
347 
348     /* DO NOT read BootMap pointer (only one, xECKD) at block #2 */
349 
350     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
351     read_block(0, sec, "Cannot read block 0 to grab boot info.");
352     if (mode == ECKD_LDL_UNLABELED) {
353         if (!magic_match(ipl1->bip.magic, ZIPL_MAGIC)) {
354             return; /* not applicable layout */
355         }
356         sclp_print("unlabeled LDL.\n");
357     }
358     verify_boot_info(&ipl1->bip);
359 
360     /* save pointer to Boot Map Table */
361     bmt_block_nr = eckd_block_num(&ipl1->bip.bp.ipl.bm_ptr.eckd.bptr.chs);
362 
363     /* save pointer to Stage1b Data */
364     s1b_block_nr = eckd_block_num(&ipl1->stage1.seek[0].chs);
365 
366     run_eckd_boot_script(bmt_block_nr, s1b_block_nr);
367     /* no return */
368 }
369 
370 static void print_eckd_msg(void)
371 {
372     char msg[] = "Using ECKD scheme (block size *****), ";
373     char *p = &msg[34], *q = &msg[30];
374     int n = virtio_get_block_size();
375 
376     /* Fill in the block size and show up the message */
377     if (n > 0 && n <= 99999) {
378         while (n) {
379             *p-- = '0' + (n % 10);
380             n /= 10;
381         }
382         while (p >= q) {
383             *p-- = ' ';
384         }
385     }
386     sclp_print(msg);
387 }
388 
389 static void ipl_eckd(void)
390 {
391     XEckdMbr *mbr = (void *)sec;
392     LDL_VTOC *vlbl = (void *)sec;
393 
394     print_eckd_msg();
395 
396     /* Grab the MBR again */
397     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
398     read_block(0, mbr, "Cannot read block 0 on DASD");
399 
400     if (magic_match(mbr->magic, IPL1_MAGIC)) {
401         ipl_eckd_cdl(); /* no return */
402     }
403 
404     /* LDL/CMS? */
405     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
406     read_block(2, vlbl, "Cannot read block 2");
407 
408     if (magic_match(vlbl->magic, CMS1_MAGIC)) {
409         ipl_eckd_ldl(ECKD_CMS); /* no return */
410     }
411     if (magic_match(vlbl->magic, LNX1_MAGIC)) {
412         ipl_eckd_ldl(ECKD_LDL); /* no return */
413     }
414 
415     ipl_eckd_ldl(ECKD_LDL_UNLABELED); /* it still may return */
416     /*
417      * Ok, it is not a LDL by any means.
418      * It still might be a CDL with zero record keys for IPL1 and IPL2
419      */
420     ipl_eckd_cdl();
421 }
422 
423 /***********************************************************************
424  * IPL a SCSI disk
425  */
426 
427 static void zipl_load_segment(ComponentEntry *entry)
428 {
429     const int max_entries = (MAX_SECTOR_SIZE / sizeof(ScsiBlockPtr));
430     ScsiBlockPtr *bprs = (void *)sec;
431     const int bprs_size = sizeof(sec);
432     block_number_t blockno;
433     uint64_t address;
434     int i;
435     char err_msg[] = "zIPL failed to read BPRS at 0xZZZZZZZZZZZZZZZZ";
436     char *blk_no = &err_msg[30]; /* where to print blockno in (those ZZs) */
437 
438     blockno = entry->data.blockno;
439     address = entry->load_address;
440 
441     debug_print_int("loading segment at block", blockno);
442     debug_print_int("addr", address);
443 
444     do {
445         memset(bprs, FREE_SPACE_FILLER, bprs_size);
446         fill_hex_val(blk_no, &blockno, sizeof(blockno));
447         read_block(blockno, bprs, err_msg);
448 
449         for (i = 0;; i++) {
450             uint64_t *cur_desc = (void *)&bprs[i];
451 
452             blockno = bprs[i].blockno;
453             if (!blockno) {
454                 break;
455             }
456 
457             /* we need the updated blockno for the next indirect entry in the
458                chain, but don't want to advance address */
459             if (i == (max_entries - 1)) {
460                 break;
461             }
462 
463             if (bprs[i].blockct == 0 && unused_space(&bprs[i + 1],
464                 sizeof(ScsiBlockPtr))) {
465                 /* This is a "continue" pointer.
466                  * This ptr is the last one in the current script section.
467                  * I.e. the next ptr must point to the unused memory area.
468                  * The blockno is not zero, so the upper loop must continue
469                  * reading next section of BPRS.
470                  */
471                 break;
472             }
473             address = virtio_load_direct(cur_desc[0], cur_desc[1], 0,
474                                          (void *)address);
475             IPL_assert(address != -1, "zIPL load segment failed");
476         }
477     } while (blockno);
478 }
479 
480 /* Run a zipl program */
481 static void zipl_run(ScsiBlockPtr *pte)
482 {
483     ComponentHeader *header;
484     ComponentEntry *entry;
485     uint8_t tmp_sec[MAX_SECTOR_SIZE];
486 
487     read_block(pte->blockno, tmp_sec, "Cannot read header");
488     header = (ComponentHeader *)tmp_sec;
489 
490     IPL_assert(magic_match(tmp_sec, ZIPL_MAGIC), "No zIPL magic in header");
491     IPL_assert(header->type == ZIPL_COMP_HEADER_IPL, "Bad header type");
492 
493     dputs("start loading images\n");
494 
495     /* Load image(s) into RAM */
496     entry = (ComponentEntry *)(&header[1]);
497     while (entry->component_type == ZIPL_COMP_ENTRY_LOAD ||
498            entry->component_type == ZIPL_COMP_ENTRY_SIGNATURE) {
499 
500         /* We don't support secure boot yet, so we skip signature entries */
501         if (entry->component_type == ZIPL_COMP_ENTRY_SIGNATURE) {
502             entry++;
503             continue;
504         }
505 
506         zipl_load_segment(entry);
507 
508         entry++;
509 
510         IPL_assert((uint8_t *)(&entry[1]) <= (tmp_sec + MAX_SECTOR_SIZE),
511                    "Wrong entry value");
512     }
513 
514     IPL_assert(entry->component_type == ZIPL_COMP_ENTRY_EXEC, "No EXEC entry");
515 
516     /* should not return */
517     jump_to_IPL_code(entry->load_address);
518 }
519 
520 static void ipl_scsi(void)
521 {
522     ScsiMbr *mbr = (void *)sec;
523     int program_table_entries = 0;
524     BootMapTable *prog_table = (void *)sec;
525     unsigned int loadparm = get_loadparm_index();
526     bool valid_entries[MAX_BOOT_ENTRIES] = {false};
527     size_t i;
528 
529     /* Grab the MBR */
530     memset(sec, FREE_SPACE_FILLER, sizeof(sec));
531     read_block(0, mbr, "Cannot read block 0");
532 
533     if (!magic_match(mbr->magic, ZIPL_MAGIC)) {
534         return;
535     }
536 
537     sclp_print("Using SCSI scheme.\n");
538     debug_print_int("MBR Version", mbr->version_id);
539     IPL_check(mbr->version_id == 1,
540               "Unknown MBR layout version, assuming version 1");
541     debug_print_int("program table", mbr->pt.blockno);
542     IPL_assert(mbr->pt.blockno, "No Program Table");
543 
544     /* Parse the program table */
545     read_block(mbr->pt.blockno, sec, "Error reading Program Table");
546     IPL_assert(magic_match(sec, ZIPL_MAGIC), "No zIPL magic in PT");
547 
548     for (i = 0; i < MAX_BOOT_ENTRIES; i++) {
549         if (prog_table->entry[i].scsi.blockno) {
550             valid_entries[i] = true;
551             program_table_entries++;
552         }
553     }
554 
555     debug_print_int("program table entries", program_table_entries);
556     IPL_assert(program_table_entries != 0, "Empty Program Table");
557 
558     if (menu_is_enabled_enum()) {
559         loadparm = menu_get_enum_boot_index(valid_entries);
560     }
561 
562     debug_print_int("loadparm", loadparm);
563     IPL_assert(loadparm < MAX_BOOT_ENTRIES, "loadparm value greater than"
564                " maximum number of boot entries allowed");
565 
566     zipl_run(&prog_table->entry[loadparm].scsi); /* no return */
567 }
568 
569 /***********************************************************************
570  * IPL El Torito ISO9660 image or DVD
571  */
572 
573 static bool is_iso_bc_entry_compatible(IsoBcSection *s)
574 {
575     uint8_t *magic_sec = (uint8_t *)(sec + ISO_SECTOR_SIZE);
576 
577     if (s->unused || !s->sector_count) {
578         return false;
579     }
580     read_iso_sector(bswap32(s->load_rba), magic_sec,
581                     "Failed to read image sector 0");
582 
583     /* Checking bytes 8 - 32 for S390 Linux magic */
584     return !memcmp(magic_sec + 8, linux_s390_magic, 24);
585 }
586 
587 /* Location of the current sector of the directory */
588 static uint32_t sec_loc[ISO9660_MAX_DIR_DEPTH];
589 /* Offset in the current sector of the directory */
590 static uint32_t sec_offset[ISO9660_MAX_DIR_DEPTH];
591 /* Remained directory space in bytes */
592 static uint32_t dir_rem[ISO9660_MAX_DIR_DEPTH];
593 
594 static inline uint32_t iso_get_file_size(uint32_t load_rba)
595 {
596     IsoVolDesc *vd = (IsoVolDesc *)sec;
597     IsoDirHdr *cur_record = &vd->vd.primary.rootdir;
598     uint8_t *temp = sec + ISO_SECTOR_SIZE;
599     int level = 0;
600 
601     read_iso_sector(ISO_PRIMARY_VD_SECTOR, sec,
602                     "Failed to read ISO primary descriptor");
603     sec_loc[0] = iso_733_to_u32(cur_record->ext_loc);
604     dir_rem[0] = 0;
605     sec_offset[0] = 0;
606 
607     while (level >= 0) {
608         IPL_assert(sec_offset[level] <= ISO_SECTOR_SIZE,
609                    "Directory tree structure violation");
610 
611         cur_record = (IsoDirHdr *)(temp + sec_offset[level]);
612 
613         if (sec_offset[level] == 0) {
614             read_iso_sector(sec_loc[level], temp,
615                             "Failed to read ISO directory");
616             if (dir_rem[level] == 0) {
617                 /* Skip self and parent records */
618                 dir_rem[level] = iso_733_to_u32(cur_record->data_len) -
619                                  cur_record->dr_len;
620                 sec_offset[level] += cur_record->dr_len;
621 
622                 cur_record = (IsoDirHdr *)(temp + sec_offset[level]);
623                 dir_rem[level] -= cur_record->dr_len;
624                 sec_offset[level] += cur_record->dr_len;
625                 continue;
626             }
627         }
628 
629         if (!cur_record->dr_len || sec_offset[level] == ISO_SECTOR_SIZE) {
630             /* Zero-padding and/or the end of current sector */
631             dir_rem[level] -= ISO_SECTOR_SIZE - sec_offset[level];
632             sec_offset[level] = 0;
633             sec_loc[level]++;
634         } else {
635             /* The directory record is valid */
636             if (load_rba == iso_733_to_u32(cur_record->ext_loc)) {
637                 return iso_733_to_u32(cur_record->data_len);
638             }
639 
640             dir_rem[level] -= cur_record->dr_len;
641             sec_offset[level] += cur_record->dr_len;
642 
643             if (cur_record->file_flags & 0x2) {
644                 /* Subdirectory */
645                 if (level == ISO9660_MAX_DIR_DEPTH - 1) {
646                     sclp_print("ISO-9660 directory depth limit exceeded\n");
647                 } else {
648                     level++;
649                     sec_loc[level] = iso_733_to_u32(cur_record->ext_loc);
650                     sec_offset[level] = 0;
651                     dir_rem[level] = 0;
652                     continue;
653                 }
654             }
655         }
656 
657         if (dir_rem[level] == 0) {
658             /* Nothing remaining */
659             level--;
660             read_iso_sector(sec_loc[level], temp,
661                             "Failed to read ISO directory");
662         }
663     }
664 
665     return 0;
666 }
667 
668 static void load_iso_bc_entry(IsoBcSection *load)
669 {
670     IsoBcSection s = *load;
671     /*
672      * According to spec, extent for each file
673      * is padded and ISO_SECTOR_SIZE bytes aligned
674      */
675     uint32_t blks_to_load = bswap16(s.sector_count) >> ET_SECTOR_SHIFT;
676     uint32_t real_size = iso_get_file_size(bswap32(s.load_rba));
677 
678     if (real_size) {
679         /* Round up blocks to load */
680         blks_to_load = (real_size + ISO_SECTOR_SIZE - 1) / ISO_SECTOR_SIZE;
681         sclp_print("ISO boot image size verified\n");
682     } else {
683         sclp_print("ISO boot image size could not be verified\n");
684     }
685 
686     read_iso_boot_image(bswap32(s.load_rba),
687                         (void *)((uint64_t)bswap16(s.load_segment)),
688                         blks_to_load);
689 
690     jump_to_low_kernel();
691 }
692 
693 static uint32_t find_iso_bc(void)
694 {
695     IsoVolDesc *vd = (IsoVolDesc *)sec;
696     uint32_t block_num = ISO_PRIMARY_VD_SECTOR;
697 
698     if (virtio_read_many(block_num++, sec, 1)) {
699         /* If primary vd cannot be read, there is no boot catalog */
700         return 0;
701     }
702 
703     while (is_iso_vd_valid(vd) && vd->type != VOL_DESC_TERMINATOR) {
704         if (vd->type == VOL_DESC_TYPE_BOOT) {
705             IsoVdElTorito *et = &vd->vd.boot;
706 
707             if (!memcmp(&et->el_torito[0], el_torito_magic, 32)) {
708                 return bswap32(et->bc_offset);
709             }
710         }
711         read_iso_sector(block_num++, sec,
712                         "Failed to read ISO volume descriptor");
713     }
714 
715     return 0;
716 }
717 
718 static IsoBcSection *find_iso_bc_entry(void)
719 {
720     IsoBcEntry *e = (IsoBcEntry *)sec;
721     uint32_t offset = find_iso_bc();
722     int i;
723     unsigned int loadparm = get_loadparm_index();
724 
725     if (!offset) {
726         return NULL;
727     }
728 
729     read_iso_sector(offset, sec, "Failed to read El Torito boot catalog");
730 
731     if (!is_iso_bc_valid(e)) {
732         /* The validation entry is mandatory */
733         panic("No valid boot catalog found!\n");
734         return NULL;
735     }
736 
737     /*
738      * Each entry has 32 bytes size, so one sector cannot contain > 64 entries.
739      * We consider only boot catalogs with no more than 64 entries.
740      */
741     for (i = 1; i < ISO_BC_ENTRY_PER_SECTOR; i++) {
742         if (e[i].id == ISO_BC_BOOTABLE_SECTION) {
743             if (is_iso_bc_entry_compatible(&e[i].body.sect)) {
744                 if (loadparm <= 1) {
745                     /* found, default, or unspecified */
746                     return &e[i].body.sect;
747                 }
748                 loadparm--;
749             }
750         }
751     }
752 
753     panic("No suitable boot entry found on ISO-9660 media!\n");
754 
755     return NULL;
756 }
757 
758 static void ipl_iso_el_torito(void)
759 {
760     IsoBcSection *s = find_iso_bc_entry();
761 
762     if (s) {
763         load_iso_bc_entry(s);
764         /* no return */
765     }
766 }
767 
768 /***********************************************************************
769  * Bus specific IPL sequences
770  */
771 
772 static void zipl_load_vblk(void)
773 {
774     if (virtio_guessed_disk_nature()) {
775         virtio_assume_iso9660();
776     }
777     ipl_iso_el_torito();
778 
779     if (virtio_guessed_disk_nature()) {
780         sclp_print("Using guessed DASD geometry.\n");
781         virtio_assume_eckd();
782     }
783     ipl_eckd();
784 }
785 
786 static void zipl_load_vscsi(void)
787 {
788     if (virtio_get_block_size() == VIRTIO_ISO_BLOCK_SIZE) {
789         /* Is it an ISO image in non-CD drive? */
790         ipl_iso_el_torito();
791     }
792 
793     sclp_print("Using guessed DASD geometry.\n");
794     virtio_assume_eckd();
795     ipl_eckd();
796 }
797 
798 /***********************************************************************
799  * IPL starts here
800  */
801 
802 void zipl_load(void)
803 {
804     VDev *vdev = virtio_get_device();
805 
806     if (vdev->is_cdrom) {
807         ipl_iso_el_torito();
808         panic("\n! Cannot IPL this ISO image !\n");
809     }
810 
811     if (virtio_get_device_type() == VIRTIO_ID_NET) {
812         jump_to_IPL_code(vdev->netboot_start_addr);
813     }
814 
815     ipl_scsi();
816 
817     switch (virtio_get_device_type()) {
818     case VIRTIO_ID_BLOCK:
819         zipl_load_vblk();
820         break;
821     case VIRTIO_ID_SCSI:
822         zipl_load_vscsi();
823         break;
824     default:
825         panic("\n! Unknown IPL device type !\n");
826     }
827 
828     panic("\n* this can never happen *\n");
829 }
830