Below is the file 'src/fs/diskd/store_dir_diskd.c' from this revision. You can also download the file.
/* * $Id: store_dir_diskd.c,v 1.58.2.12 2005/05/01 10:48:07 serassio Exp $ * * DEBUG: section 47 Store Directory Routines * AUTHOR: Duane Wessels * * SQUID Web Proxy Cache http://www.squid-cache.org/ * ---------------------------------------------------------- * * Squid is the result of efforts by numerous individuals from * the Internet community; see the CONTRIBUTORS file for full * details. Many organizations have provided support for Squid's * development; see the SPONSORS file for full details. Squid is * Copyrighted (C) 2001 by the Regents of the University of * California; see the COPYRIGHT file for full details. Squid * incorporates software developed and/or copyrighted by other * sources; see the CREDITS file for full details. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA. * */ #include "squid.h" #include <sys/ipc.h> #include <sys/msg.h> #include <sys/shm.h> #include "store_diskd.h" #define DefaultLevelOneDirs 16 #define DefaultLevelTwoDirs 256 #define STORE_META_BDISKDZ 4096 diskd_stats_t diskd_stats; typedef struct _RebuildState RebuildState; struct _RebuildState { SwapDir *sd; int n_read; FILE *log; int speed; int curlvl1; int curlvl2; struct { unsigned int need_to_validate:1; unsigned int clean:1; unsigned int init:1; } flags; int done; int in_dir; int fn; struct dirent *entry; DIR *td; char fullpath[SQUID_MAXPATHLEN]; char fullfilename[SQUID_MAXPATHLEN]; struct _store_rebuild_data counts; }; static int n_diskd_dirs = 0; static int *diskd_dir_index = NULL; MemPool *diskd_state_pool = NULL; static int diskd_initialised = 0; static char *storeDiskdDirSwapSubDir(SwapDir *, int subdirn); static int storeDiskdDirCreateDirectory(const char *path, int); static int storeDiskdDirVerifyCacheDirs(SwapDir *); static int storeDiskdDirVerifyDirectory(const char *path); static void storeDiskdDirCreateSwapSubDirs(SwapDir *); static char *storeDiskdDirSwapLogFile(SwapDir *, const char *); static EVH storeDiskdDirRebuildFromDirectory; static EVH storeDiskdDirRebuildFromSwapLog; static int storeDiskdDirGetNextFile(RebuildState *, sfileno *, int *size); static StoreEntry *storeDiskdDirAddDiskRestore(SwapDir * SD, const cache_key * key, sfileno file_number, squid_file_sz swap_file_sz, time_t expires, time_t timestamp, time_t lastref, time_t lastmod, u_num32 refcount, u_short flags, int clean); static void storeDiskdDirRebuild(SwapDir * sd); static void storeDiskdDirCloseTmpSwapLog(SwapDir * sd); static FILE *storeDiskdDirOpenTmpSwapLog(SwapDir *, int *, int *); static STLOGOPEN storeDiskdDirOpenSwapLog; static STINIT storeDiskdDirInit; static STFREE storeDiskdDirFree; static STLOGCLEANSTART storeDiskdDirWriteCleanStart; static STLOGCLEANNEXTENTRY storeDiskdDirCleanLogNextEntry; static STLOGCLEANWRITE storeDiskdDirWriteCleanEntry; static STLOGCLEANDONE storeDiskdDirWriteCleanDone; static STLOGCLOSE storeDiskdDirCloseSwapLog; static STLOGWRITE storeDiskdDirSwapLog; static STNEWFS storeDiskdDirNewfs; static STDUMP storeDiskdDirDump; static STMAINTAINFS storeDiskdDirMaintain; static STCHECKOBJ storeDiskdDirCheckObj; static STCHECKLOADAV storeDiskdDirCheckLoadAv; static STREFOBJ storeDiskdDirRefObj; static STUNREFOBJ storeDiskdDirUnrefObj; static QS rev_int_sort; static int storeDiskdDirClean(int swap_index); static EVH storeDiskdDirCleanEvent; static int storeDiskdDirIs(SwapDir * sd); static int storeDiskdFilenoBelongsHere(int fn, int F0, int F1, int F2); static int storeDiskdCleanupDoubleCheck(SwapDir *, StoreEntry *); static void storeDiskdDirStats(SwapDir *, StoreEntry *); static void storeDiskdDirInitBitmap(SwapDir *); static int storeDiskdDirValidFileno(SwapDir *, sfileno, int); static void storeDiskdStats(StoreEntry * sentry); static void storeDiskdDirSync(SwapDir *); /* The only externally visible interface */ STSETUP storeFsSetup_diskd; /* * These functions were ripped straight out of the heart of store_dir.c. * They assume that the given filenum is on a diskd partiton, which may or * may not be true.. * XXX this evilness should be tidied up at a later date! */ static int storeDiskdDirMapBitTest(SwapDir * SD, sfileno filn) { diskdinfo_t *diskdinfo; diskdinfo = SD->fsdata; return file_map_bit_test(diskdinfo->map, filn); } static void storeDiskdDirMapBitSet(SwapDir * SD, sfileno filn) { diskdinfo_t *diskdinfo; diskdinfo = SD->fsdata; file_map_bit_set(diskdinfo->map, filn); } void storeDiskdDirMapBitReset(SwapDir * SD, sfileno filn) { diskdinfo_t *diskdinfo; diskdinfo = SD->fsdata; /* * We have to test the bit before calling file_map_bit_reset. * file_map_bit_reset doesn't do bounds checking. It assumes * filn is a valid file number, but it might not be because * the map is dynamic in size. Also clearing an already clear * bit puts the map counter of-of-whack. */ if (file_map_bit_test(diskdinfo->map, filn)) file_map_bit_reset(diskdinfo->map, filn); } int storeDiskdDirMapBitAllocate(SwapDir * SD) { diskdinfo_t *diskdinfo = SD->fsdata; int fn; fn = file_map_allocate(diskdinfo->map, diskdinfo->suggest); file_map_bit_set(diskdinfo->map, fn); diskdinfo->suggest = fn + 1; return fn; } /* * Initialise the diskd bitmap * * If there already is a bitmap, and the numobjects is larger than currently * configured, we allocate a new bitmap and 'grow' the old one into it. */ static void storeDiskdDirInitBitmap(SwapDir * sd) { diskdinfo_t *diskdinfo = sd->fsdata; if (diskdinfo->map == NULL) { /* First time */ diskdinfo->map = file_map_create(); } else if (diskdinfo->map->max_n_files) { /* it grew, need to expand */ /* XXX We don't need it anymore .. */ } /* else it shrunk, and we leave the old one in place */ } static char * storeDiskdDirSwapSubDir(SwapDir * sd, int subdirn) { diskdinfo_t *diskdinfo = sd->fsdata; LOCAL_ARRAY(char, fullfilename, SQUID_MAXPATHLEN); assert(0 <= subdirn && subdirn < diskdinfo->l1); snprintf(fullfilename, SQUID_MAXPATHLEN, "%s/%02X", sd->path, subdirn); return fullfilename; } static int storeDiskdDirCreateDirectory(const char *path, int should_exist) { int created = 0; struct stat st; getCurrentTime(); if (0 == stat(path, &st)) { if (S_ISDIR(st.st_mode)) { debug(20, should_exist ? 3 : 1) ("%s exists\n", path); } else { fatalf("Swap directory %s is not a directory.", path); } } else if (0 == mkdir(path, 0755)) { debug(20, should_exist ? 1 : 3) ("%s created\n", path); created = 1; } else { fatalf("Failed to make swap directory %s: %s", path, xstrerror()); } return created; } static int storeDiskdDirVerifyDirectory(const char *path) { struct stat sb; if (stat(path, &sb) < 0) { debug(20, 0) ("%s: %s\n", path, xstrerror()); return -1; } if (S_ISDIR(sb.st_mode) == 0) { debug(20, 0) ("%s is not a directory\n", path); return -1; } return 0; } /* * This function is called by storeDiskdDirInit(). If this returns < 0, * then Squid exits, complains about swap directories not * existing, and instructs the admin to run 'squid -z' */ static int storeDiskdDirVerifyCacheDirs(SwapDir * sd) { diskdinfo_t *diskdinfo = sd->fsdata; int j; const char *path = sd->path; if (storeDiskdDirVerifyDirectory(path) < 0) return -1; for (j = 0; j < diskdinfo->l1; j++) { path = storeDiskdDirSwapSubDir(sd, j); if (storeDiskdDirVerifyDirectory(path) < 0) return -1; } return 0; } static void storeDiskdDirCreateSwapSubDirs(SwapDir * sd) { diskdinfo_t *diskdinfo = sd->fsdata; int i, k; int should_exist; LOCAL_ARRAY(char, name, MAXPATHLEN); for (i = 0; i < diskdinfo->l1; i++) { snprintf(name, MAXPATHLEN, "%s/%02X", sd->path, i); if (storeDiskdDirCreateDirectory(name, 0)) should_exist = 0; else should_exist = 1; debug(47, 1) ("Making directories in %s\n", name); for (k = 0; k < diskdinfo->l2; k++) { snprintf(name, MAXPATHLEN, "%s/%02X/%02X", sd->path, i, k); storeDiskdDirCreateDirectory(name, should_exist); } } } static char * storeDiskdDirSwapLogFile(SwapDir * sd, const char *ext) { LOCAL_ARRAY(char, path, SQUID_MAXPATHLEN); LOCAL_ARRAY(char, pathtmp, SQUID_MAXPATHLEN); LOCAL_ARRAY(char, digit, 32); char *pathtmp2; if (Config.Log.swap) { xstrncpy(pathtmp, sd->path, SQUID_MAXPATHLEN - 64); pathtmp2 = pathtmp; while ((pathtmp2 = strchr(pathtmp2, '/')) != NULL) *pathtmp2 = '.'; while (strlen(pathtmp) && pathtmp[strlen(pathtmp) - 1] == '.') pathtmp[strlen(pathtmp) - 1] = '\0'; for (pathtmp2 = pathtmp; *pathtmp2 == '.'; pathtmp2++); snprintf(path, SQUID_MAXPATHLEN - 64, Config.Log.swap, pathtmp2); if (strncmp(path, Config.Log.swap, SQUID_MAXPATHLEN - 64) == 0) { strcat(path, "."); snprintf(digit, 32, "%02d", sd->index); strncat(path, digit, 3); } } else { xstrncpy(path, sd->path, SQUID_MAXPATHLEN - 64); strcat(path, "/swap.state"); } if (ext) strncat(path, ext, 16); return path; } static void storeDiskdDirOpenSwapLog(SwapDir * sd) { diskdinfo_t *diskdinfo = sd->fsdata; char *path; int fd; path = storeDiskdDirSwapLogFile(sd, NULL); fd = file_open(path, O_WRONLY | O_CREAT); if (fd < 0) { debug(50, 1) ("%s: %s\n", path, xstrerror()); fatal("storeDiskdDirOpenSwapLog: Failed to open swap log."); } debug(47, 3) ("Cache Dir #%d log opened on FD %d\n", sd->index, fd); diskdinfo->swaplog_fd = fd; if (0 == n_diskd_dirs) assert(NULL == diskd_dir_index); n_diskd_dirs++; assert(n_diskd_dirs <= Config.cacheSwap.n_configured); } static void storeDiskdDirCloseSwapLog(SwapDir * sd) { diskdinfo_t *diskdinfo = sd->fsdata; if (diskdinfo->swaplog_fd < 0) /* not open */ return; file_close(diskdinfo->swaplog_fd); debug(47, 3) ("Cache Dir #%d log closed on FD %d\n", sd->index, diskdinfo->swaplog_fd); diskdinfo->swaplog_fd = -1; n_diskd_dirs--; assert(n_diskd_dirs >= 0); if (0 == n_diskd_dirs) safe_free(diskd_dir_index); } static void storeDiskdDirInit(SwapDir * sd) { static int started_clean_event = 0; int x; int i; int rfd; int ikey; const char *args[5]; char skey1[32]; char skey2[32]; char skey3[32]; diskdinfo_t *diskdinfo = sd->fsdata; static const char *errmsg = "\tFailed to verify one of the swap directories, Check cache.log\n" "\tfor details. Run 'squid -z' to create swap directories\n" "\tif needed, or if running Squid for the first time."; ikey = (getpid() << 10) + (sd->index << 2); ikey &= 0x7fffffff; diskdinfo->smsgid = msgget((key_t) ikey, 0700 | IPC_CREAT); if (diskdinfo->smsgid < 0) { debug(50, 0) ("storeDiskdInit: msgget: %s\n", xstrerror()); fatal("msgget failed"); } diskdinfo->rmsgid = msgget((key_t) (ikey + 1), 0700 | IPC_CREAT); if (diskdinfo->rmsgid < 0) { debug(50, 0) ("storeDiskdInit: msgget: %s\n", xstrerror()); fatal("msgget failed"); } diskdinfo->shm.id = shmget((key_t) (ikey + 2), SHMBUFS * SHMBUF_BLKSZ, 0600 | IPC_CREAT); if (diskdinfo->shm.id < 0) { debug(50, 0) ("storeDiskdInit: shmget: %s\n", xstrerror()); fatal("shmget failed"); } diskdinfo->shm.buf = shmat(diskdinfo->shm.id, NULL, 0); if (diskdinfo->shm.buf == (void *) -1) { debug(50, 0) ("storeDiskdInit: shmat: %s\n", xstrerror()); fatal("shmat failed"); } diskdinfo->shm.inuse_map = xcalloc((SHMBUFS + 7) / 8, 1); diskd_stats.shmbuf_count += SHMBUFS; for (i = 0; i < SHMBUFS; i++) { CBIT_SET(diskdinfo->shm.inuse_map, i); storeDiskdShmPut(sd, i * SHMBUF_BLKSZ); } snprintf(skey1, 32, "%d", ikey); snprintf(skey2, 32, "%d", ikey + 1); snprintf(skey3, 32, "%d", ikey + 2); args[0] = "diskd"; args[1] = skey1; args[2] = skey2; args[3] = skey3; args[4] = NULL; #if HAVE_POLL && defined(_SQUID_OSF_) /* pipes and poll() don't get along on DUNIX -DW */ x = ipcCreate(IPC_TCP_SOCKET, #else x = ipcCreate(IPC_FIFO, #endif Config.Program.diskd, args, "diskd", &rfd, &diskdinfo->wfd); if (x < 0) fatalf("execl: %s", Config.Program.diskd); if (rfd != diskdinfo->wfd) comm_close(rfd); fd_note(diskdinfo->wfd, "squid -> diskd"); commSetTimeout(diskdinfo->wfd, -1, NULL, NULL); commSetNonBlocking(diskdinfo->wfd); storeDiskdDirInitBitmap(sd); if (storeDiskdDirVerifyCacheDirs(sd) < 0) fatal(errmsg); storeDiskdDirOpenSwapLog(sd); storeDiskdDirRebuild(sd); if (!started_clean_event) { eventAdd("storeDirClean", storeDiskdDirCleanEvent, NULL, 15.0, 1); started_clean_event = 1; } (void) storeDirGetBlkSize(sd->path, &sd->fs.blksize); comm_quick_poll_required(); } static void storeDiskdStats(StoreEntry * sentry) { storeAppendPrintf(sentry, "sent_count: %d\n", diskd_stats.sent_count); storeAppendPrintf(sentry, "recv_count: %d\n", diskd_stats.recv_count); storeAppendPrintf(sentry, "max_away: %d\n", diskd_stats.max_away); storeAppendPrintf(sentry, "max_shmuse: %d\n", diskd_stats.max_shmuse); storeAppendPrintf(sentry, "open_fail_queue_len: %d\n", diskd_stats.open_fail_queue_len); storeAppendPrintf(sentry, "block_queue_len: %d\n", diskd_stats.block_queue_len); diskd_stats.max_away = diskd_stats.max_shmuse = 0; storeAppendPrintf(sentry, "\n OPS SUCCESS FAIL\n"); storeAppendPrintf(sentry, "%7s %9d %9d %7d\n", "open", diskd_stats.open.ops, diskd_stats.open.success, diskd_stats.open.fail); storeAppendPrintf(sentry, "%7s %9d %9d %7d\n", "create", diskd_stats.create.ops, diskd_stats.create.success, diskd_stats.create.fail); storeAppendPrintf(sentry, "%7s %9d %9d %7d\n", "close", diskd_stats.close.ops, diskd_stats.close.success, diskd_stats.close.fail); storeAppendPrintf(sentry, "%7s %9d %9d %7d\n", "unlink", diskd_stats.unlink.ops, diskd_stats.unlink.success, diskd_stats.unlink.fail); storeAppendPrintf(sentry, "%7s %9d %9d %7d\n", "read", diskd_stats.read.ops, diskd_stats.read.success, diskd_stats.read.fail); storeAppendPrintf(sentry, "%7s %9d %9d %7d\n", "write", diskd_stats.write.ops, diskd_stats.write.success, diskd_stats.write.fail); } /* * storeDiskdDirSync * * Sync any pending data. We just sit around and read the queue * until the data has finished writing. */ static void storeDiskdDirSync(SwapDir * SD) { static time_t lastmsg = 0; diskdinfo_t *diskdinfo = SD->fsdata; while (diskdinfo->away > 0) { if (squid_curtime > lastmsg) { debug(47, 1) ("storeDiskdDirSync: %d messages away\n", diskdinfo->away); lastmsg = squid_curtime; } storeDiskdDirCallback(SD); } } /* * storeDiskdDirCallback * * Handle callbacks. If we have more than magic2 requests away, we block * until the queue is below magic2. Otherwise, we simply return when we * don't get a message. */ int storeDiskdDirCallback(SwapDir * SD) { diomsg M; int x; diskdinfo_t *diskdinfo = SD->fsdata; int retval = 0; if (diskdinfo->away >= diskdinfo->magic2) { diskd_stats.block_queue_len++; retval = 1; /* We might not have anything to do, but our queue * is full.. */ } if (diskd_stats.sent_count - diskd_stats.recv_count > diskd_stats.max_away) { diskd_stats.max_away = diskd_stats.sent_count - diskd_stats.recv_count; } while (1) { memset(&M, '\0', sizeof(M)); x = msgrcv(diskdinfo->rmsgid, &M, msg_snd_rcv_sz, 0, IPC_NOWAIT); if (x < 0) break; else if (x != msg_snd_rcv_sz) { debug(79, 1) ("storeDiskdDirCallback: msgget returns %d\n", x); break; } diskd_stats.recv_count++; diskdinfo->away--; storeDiskdHandle(&M); retval = 1; /* Return that we've actually done some work */ if (M.shm_offset > -1) storeDiskdShmPut(SD, M.shm_offset); } return retval; } static void storeDiskdDirRebuildComplete(RebuildState * rb) { if (rb->log) { debug(47, 1) ("Done reading %s swaplog (%d entries)\n", rb->sd->path, rb->n_read); fclose(rb->log); rb->log = NULL; } else { debug(47, 1) ("Done scanning %s (%d entries)\n", rb->sd->path, rb->counts.scancount); } store_dirs_rebuilding--; storeDiskdDirCloseTmpSwapLog(rb->sd); storeRebuildComplete(&rb->counts); cbdataFree(rb); } static void storeDiskdDirRebuildFromDirectory(void *data) { RebuildState *rb = data; SwapDir *SD = rb->sd; LOCAL_ARRAY(char, hdr_buf, SM_PAGE_SIZE); StoreEntry *e = NULL; StoreEntry tmpe; cache_key key[MD5_DIGEST_CHARS]; sfileno filn = 0; int count; int size; struct stat sb; int swap_hdr_len; int fd = -1; tlv *tlv_list; tlv *t; assert(rb != NULL); debug(20, 3) ("storeDiskdDirRebuildFromDirectory: DIR #%d\n", rb->sd->index); for (count = 0; count < rb->speed; count++) { assert(fd == -1); fd = storeDiskdDirGetNextFile(rb, &filn, &size); if (fd == -2) { storeDiskdDirRebuildComplete(rb); return; } else if (fd < 0) { continue; } assert(fd > -1); /* lets get file stats here */ if (fstat(fd, &sb) < 0) { debug(20, 1) ("storeDiskdDirRebuildFromDirectory: fstat(FD %d): %s\n", fd, xstrerror()); file_close(fd); store_open_disk_fd--; fd = -1; continue; } if ((++rb->counts.scancount & 0xFFFF) == 0) debug(20, 3) (" %s %7d files opened so far.\n", rb->sd->path, rb->counts.scancount); debug(20, 9) ("file_in: fd=%d %08X\n", fd, filn); statCounter.syscalls.disk.reads++; if (FD_READ_METHOD(fd, hdr_buf, SM_PAGE_SIZE) < 0) { debug(20, 1) ("storeDiskdDirRebuildFromDirectory: read(FD %d): %s\n", fd, xstrerror()); file_close(fd); store_open_disk_fd--; fd = -1; continue; } file_close(fd); store_open_disk_fd--; fd = -1; swap_hdr_len = 0; #if USE_TRUNCATE if (sb.st_size == 0) continue; #endif tlv_list = storeSwapMetaUnpack(hdr_buf, &swap_hdr_len); if (tlv_list == NULL) { debug(20, 1) ("storeDiskdDirRebuildFromDirectory: failed to get meta data\n"); /* XXX shouldn't this be a call to storeDiskdUnlink ? */ storeDiskdDirUnlinkFile(SD, filn); continue; } debug(20, 3) ("storeDiskdDirRebuildFromDirectory: successful swap meta unpacking\n"); memset(key, '\0', MD5_DIGEST_CHARS); memset(&tmpe, '\0', sizeof(StoreEntry)); for (t = tlv_list; t; t = t->next) { switch (t->type) { case STORE_META_KEY: assert(t->length == MD5_DIGEST_CHARS); xmemcpy(key, t->value, MD5_DIGEST_CHARS); break; #if SIZEOF_SQUID_FILE_SZ == SIZEOF_SIZE_T case STORE_META_STD: assert(t->length == STORE_HDR_METASIZE); xmemcpy(&tmpe.timestamp, t->value, STORE_HDR_METASIZE); break; #else case STORE_META_STD_LFS: assert(t->length == STORE_HDR_METASIZE); xmemcpy(&tmpe.timestamp, t->value, STORE_HDR_METASIZE); break; case STORE_META_STD: assert(t->length == STORE_HDR_METASIZE_OLD); { struct { time_t timestamp; time_t lastref; time_t expires; time_t lastmod; size_t swap_file_sz; u_short refcount; u_short flags; } *tmp = t->value; assert(sizeof(*tmp) == STORE_HDR_METASIZE_OLD); tmpe.timestamp = tmp->timestamp; tmpe.lastref = tmp->lastref; tmpe.expires = tmp->expires; tmpe.lastmod = tmp->lastmod; tmpe.swap_file_sz = tmp->swap_file_sz; tmpe.refcount = tmp->refcount; tmpe.flags = tmp->flags; } break; #endif default: break; } } storeSwapTLVFree(tlv_list); tlv_list = NULL; if (storeKeyNull(key)) { debug(20, 1) ("storeDiskdDirRebuildFromDirectory: NULL key\n"); storeDiskdDirUnlinkFile(SD, filn); continue; } tmpe.hash.key = key; /* check sizes */ if (tmpe.swap_file_sz == 0) { tmpe.swap_file_sz = sb.st_size; } else if (tmpe.swap_file_sz == sb.st_size - swap_hdr_len) { tmpe.swap_file_sz = sb.st_size; } else if (tmpe.swap_file_sz != sb.st_size) { debug(20, 1) ("storeDiskdDirRebuildFromDirectory: SIZE MISMATCH %ld!=%ld\n", (long int) tmpe.swap_file_sz, (long int) sb.st_size); storeDiskdDirUnlinkFile(SD, filn); continue; } if (EBIT_TEST(tmpe.flags, KEY_PRIVATE)) { storeDiskdDirUnlinkFile(SD, filn); rb->counts.badflags++; continue; } e = storeGet(key); if (e && e->lastref >= tmpe.lastref) { /* key already exists, current entry is newer */ /* keep old, ignore new */ rb->counts.dupcount++; continue; } else if (NULL != e) { /* URL already exists, this swapfile not being used */ /* junk old, load new */ storeRelease(e); /* release old entry */ rb->counts.dupcount++; } rb->counts.objcount++; storeEntryDump(&tmpe, 5); e = storeDiskdDirAddDiskRestore(SD, key, filn, tmpe.swap_file_sz, tmpe.expires, tmpe.timestamp, tmpe.lastref, tmpe.lastmod, tmpe.refcount, /* refcount */ tmpe.flags, /* flags */ (int) rb->flags.clean); storeDirSwapLog(e, SWAP_LOG_ADD); } eventAdd("storeRebuild", storeDiskdDirRebuildFromDirectory, rb, 0.0, 1); } static void storeDiskdDirRebuildFromSwapLog(void *data) { RebuildState *rb = data; SwapDir *SD = rb->sd; StoreEntry *e = NULL; storeSwapLogData s; size_t ss = sizeof(storeSwapLogData); int count; int used; /* is swapfile already in use? */ int disk_entry_newer; /* is the log entry newer than current entry? */ double x; assert(rb != NULL); /* load a number of objects per invocation */ for (count = 0; count < rb->speed; count++) { if (fread(&s, ss, 1, rb->log) != 1) { storeDiskdDirRebuildComplete(rb); return; } rb->n_read++; /* * BC: during 2.4 development, we changed the way swap file * numbers are assigned and stored. The high 16 bits used * to encode the SD index number. There used to be a call * to storeDirProperFileno here that re-assigned the index * bits. Now, for backwards compatibility, we just need * to mask it off. */ s.swap_filen &= 0x00FFFFFF; debug(20, 3) ("storeDiskdDirRebuildFromSwapLog: %s %s %08X\n", swap_log_op_str[(int) s.op], storeKeyText(s.key), s.swap_filen); if (s.op == SWAP_LOG_ADD) { /* * Here we have some special checks for large files. * I've been seeing a system crash followed by a reboot * that seems to corrupt the swap log. Squid believes * that the disk holds some really large files. It * complains about using being over the high water mark * and proceeds to delete files as fast as it can. To * prevent that, we call stat() on sufficiently large * files (>128KB) and reject those that are missing or * have the wrong size. */ struct stat sb; char *p = storeDiskdDirFullPath(SD, s.swap_filen, NULL); if (s.swap_file_sz < (1 << 17)) { (void) 0; } else if (stat(p, &sb) < 0) { debug(47, 2) ("its missing!: %s\n", p); continue; } else if (sb.st_size != s.swap_file_sz) { debug(47, 2) ("size mismatch!: stat=%d, log=%d\n", (int) sb.st_size, (int) s.swap_file_sz); continue; } else { debug(47, 2) ("big file (%d bytes) checks out\n", (int) s.swap_file_sz); } } else if (s.op == SWAP_LOG_DEL) { /* Delete unless we already have a newer copy */ if ((e = storeGet(s.key)) != NULL && s.lastref > e->lastref) { /* * Make sure we don't unlink the file, it might be * in use by a subsequent entry. Also note that * we don't have to subtract from store_swap_size * because adding to store_swap_size happens in * the cleanup procedure. */ storeExpireNow(e); storeReleaseRequest(e); if (e->swap_filen > -1) { storeDiskdDirReplRemove(e); storeDiskdDirMapBitReset(SD, e->swap_filen); e->swap_filen = -1; e->swap_dirn = -1; } storeRelease(e); rb->counts.objcount--; rb->counts.cancelcount++; } continue; } else { x = log(++rb->counts.bad_log_op) / log(10.0); if (0.0 == x - (double) (int) x) debug(20, 1) ("WARNING: %d invalid swap log entries found\n", rb->counts.bad_log_op); rb->counts.invalid++; continue; } if ((++rb->counts.scancount & 0xFFF) == 0) { struct stat sb; if (0 == fstat(fileno(rb->log), &sb)) storeRebuildProgress(SD->index, (int) sb.st_size / ss, rb->n_read); } if (!storeDiskdDirValidFileno(SD, s.swap_filen, 0)) { rb->counts.invalid++; continue; } if (EBIT_TEST(s.flags, KEY_PRIVATE)) { rb->counts.badflags++; continue; } e = storeGet(s.key); used = storeDiskdDirMapBitTest(SD, s.swap_filen); /* If this URL already exists in the cache, does the swap log * appear to have a newer entry? Compare 'lastref' from the * swap log to e->lastref. */ disk_entry_newer = e ? (s.lastref > e->lastref ? 1 : 0) : 0; if (used && !disk_entry_newer) { /* log entry is old, ignore it */ rb->counts.clashcount++; continue; } else if (used && e && e->swap_filen == s.swap_filen && e->swap_dirn == SD->index) { /* swapfile taken, same URL, newer, update meta */ if (e->store_status == STORE_OK) { e->lastref = s.timestamp; e->timestamp = s.timestamp; e->expires = s.expires; e->lastmod = s.lastmod; e->flags = s.flags; e->refcount += s.refcount; storeDiskdDirUnrefObj(SD, e); } else { debug_trap("storeDiskdDirRebuildFromSwapLog: bad condition"); debug(20, 1) ("\tSee %s:%d\n", __FILE__, __LINE__); } continue; } else if (used) { /* swapfile in use, not by this URL, log entry is newer */ /* This is sorta bad: the log entry should NOT be newer at this * point. If the log is dirty, the filesize check should have * caught this. If the log is clean, there should never be a * newer entry. */ debug(20, 1) ("WARNING: newer swaplog entry for dirno %d, fileno %08X\n", SD->index, s.swap_filen); /* I'm tempted to remove the swapfile here just to be safe, * but there is a bad race condition in the NOVM version if * the swapfile has recently been opened for writing, but * not yet opened for reading. Because we can't map * swapfiles back to StoreEntrys, we don't know the state * of the entry using that file. */ /* We'll assume the existing entry is valid, probably because * were in a slow rebuild and the the swap file number got taken * and the validation procedure hasn't run. */ assert(rb->flags.need_to_validate); rb->counts.clashcount++; continue; } else if (e && !disk_entry_newer) { /* key already exists, current entry is newer */ /* keep old, ignore new */ rb->counts.dupcount++; continue; } else if (e) { /* key already exists, this swapfile not being used */ /* junk old, load new */ storeExpireNow(e); storeReleaseRequest(e); if (e->swap_filen > -1) { storeDiskdDirReplRemove(e); /* Make sure we don't actually unlink the file */ storeDiskdDirMapBitReset(SD, e->swap_filen); e->swap_filen = -1; e->swap_dirn = -1; } storeRelease(e); rb->counts.dupcount++; } else { /* URL doesnt exist, swapfile not in use */ /* load new */ (void) 0; } /* update store_swap_size */ rb->counts.objcount++; e = storeDiskdDirAddDiskRestore(SD, s.key, s.swap_filen, s.swap_file_sz, s.expires, s.timestamp, s.lastref, s.lastmod, s.refcount, s.flags, (int) rb->flags.clean); storeDirSwapLog(e, SWAP_LOG_ADD); } eventAdd("storeRebuild", storeDiskdDirRebuildFromSwapLog, rb, 0.0, 1); } #if SIZEOF_SQUID_FILE_SZ != SIZEOF_SIZE_T /* This is an exact copy of the above, but using storeSwapLogDataOld entry type */ static void storeDiskdDirRebuildFromSwapLogOld(void *data) { RebuildState *rb = data; SwapDir *SD = rb->sd; StoreEntry *e = NULL; storeSwapLogDataOld s; size_t ss = sizeof(storeSwapLogDataOld); int count; int used; /* is swapfile already in use? */ int disk_entry_newer; /* is the log entry newer than current entry? */ double x; assert(rb != NULL); /* load a number of objects per invocation */ for (count = 0; count < rb->speed; count++) { if (fread(&s, ss, 1, rb->log) != 1) { storeDiskdDirRebuildComplete(rb); return; } rb->n_read++; /* * BC: during 2.4 development, we changed the way swap file * numbers are assigned and stored. The high 16 bits used * to encode the SD index number. There used to be a call * to storeDirProperFileno here that re-assigned the index * bits. Now, for backwards compatibility, we just need * to mask it off. */ s.swap_filen &= 0x00FFFFFF; debug(20, 3) ("storeDiskdDirRebuildFromSwapLog: %s %s %08X\n", swap_log_op_str[(int) s.op], storeKeyText(s.key), s.swap_filen); if (s.op == SWAP_LOG_ADD) { /* * Here we have some special checks for large files. * I've been seeing a system crash followed by a reboot * that seems to corrupt the swap log. Squid believes * that the disk holds some really large files. It * complains about using being over the high water mark * and proceeds to delete files as fast as it can. To * prevent that, we call stat() on sufficiently large * files (>128KB) and reject those that are missing or * have the wrong size. */ struct stat sb; char *p = storeDiskdDirFullPath(SD, s.swap_filen, NULL); if (s.swap_file_sz < (1 << 17)) { (void) 0; } else if (stat(p, &sb) < 0) { debug(47, 2) ("its missing!: %s\n", p); continue; } else if (sb.st_size != s.swap_file_sz) { debug(47, 2) ("size mismatch!: stat=%d, log=%d\n", (int) sb.st_size, (int) s.swap_file_sz); continue; } else { debug(47, 2) ("big file (%d bytes) checks out\n", (int) s.swap_file_sz); } } else if (s.op == SWAP_LOG_DEL) { /* Delete unless we already have a newer copy */ if ((e = storeGet(s.key)) != NULL && s.lastref > e->lastref) { /* * Make sure we don't unlink the file, it might be * in use by a subsequent entry. Also note that * we don't have to subtract from store_swap_size * because adding to store_swap_size happens in * the cleanup procedure. */ storeExpireNow(e); storeReleaseRequest(e); if (e->swap_filen > -1) { storeDiskdDirReplRemove(e); storeDiskdDirMapBitReset(SD, e->swap_filen); e->swap_filen = -1; e->swap_dirn = -1; } storeRelease(e); rb->counts.objcount--; rb->counts.cancelcount++; } continue; } else { x = log(++rb->counts.bad_log_op) / log(10.0); if (0.0 == x - (double) (int) x) debug(20, 1) ("WARNING: %d invalid swap log entries found\n", rb->counts.bad_log_op); rb->counts.invalid++; continue; } if ((++rb->counts.scancount & 0xFFF) == 0) { struct stat sb; if (0 == fstat(fileno(rb->log), &sb)) storeRebuildProgress(SD->index, (int) sb.st_size / ss, rb->n_read); } if (!storeDiskdDirValidFileno(SD, s.swap_filen, 0)) { rb->counts.invalid++; continue; } if (EBIT_TEST(s.flags, KEY_PRIVATE)) { rb->counts.badflags++; continue; } e = storeGet(s.key); used = storeDiskdDirMapBitTest(SD, s.swap_filen); /* If this URL already exists in the cache, does the swap log * appear to have a newer entry? Compare 'lastref' from the * swap log to e->lastref. */ disk_entry_newer = e ? (s.lastref > e->lastref ? 1 : 0) : 0; if (used && !disk_entry_newer) { /* log entry is old, ignore it */ rb->counts.clashcount++; continue; } else if (used && e && e->swap_filen == s.swap_filen && e->swap_dirn == SD->index) { /* swapfile taken, same URL, newer, update meta */ if (e->store_status == STORE_OK) { e->lastref = s.timestamp; e->timestamp = s.timestamp; e->expires = s.expires; e->lastmod = s.lastmod; e->flags = s.flags; e->refcount += s.refcount; storeDiskdDirUnrefObj(SD, e); } else { debug_trap("storeDiskdDirRebuildFromSwapLog: bad condition"); debug(20, 1) ("\tSee %s:%d\n", __FILE__, __LINE__); } continue; } else if (used) { /* swapfile in use, not by this URL, log entry is newer */ /* This is sorta bad: the log entry should NOT be newer at this * point. If the log is dirty, the filesize check should have * caught this. If the log is clean, there should never be a * newer entry. */ debug(20, 1) ("WARNING: newer swaplog entry for dirno %d, fileno %08X\n", SD->index, s.swap_filen); /* I'm tempted to remove the swapfile here just to be safe, * but there is a bad race condition in the NOVM version if * the swapfile has recently been opened for writing, but * not yet opened for reading. Because we can't map * swapfiles back to StoreEntrys, we don't know the state * of the entry using that file. */ /* We'll assume the existing entry is valid, probably because * were in a slow rebuild and the the swap file number got taken * and the validation procedure hasn't run. */ assert(rb->flags.need_to_validate); rb->counts.clashcount++; continue; } else if (e && !disk_entry_newer) { /* key already exists, current entry is newer */ /* keep old, ignore new */ rb